workmux
Open-source Mac and Linux app. Agents run on your computer or on your own server.
Product websiteworkmux.dev Compare with others Missing or wrong data?
- Desktop app
- Mac and Linux
- Browser
- No
- Phone
- No
- Command line
- Yes
- Agents
- Claude Code, Codex, Gemini CLI, OpenCode, Pi, Grok and Antigravity
- Your subscription
- Works with your own plan
- Agents run
- On your computer and on your own server
- Laptop closed
- Keeps going
- Isolation
- Worktree, Container and VM
- Parallel agents
- Yes
- Scheduled runs
- No
- Coordinator agent
- Yes
- Built-in issues
- No
- Pull requests
- Yes
- Review screen
- Yes
- Team
- No
- Self-hosting
- Yes
- Licence
- MIT
- Price from
- Free
- How it charges
- No paid plan and bring your own key
- Paid plans from
- Not listed
Receipts
- Checked
- Findings
- 50 published · 48 cross-checked by a second agent · read them all
- Sources
- 45 public pages on 4 sites: workmux.dev, github.com, api.github.com, raw.githubusercontent.com
- GitHub stars
- 3,037 on · api.github.com · repository attention, not users or capability
- Screenshot
- workmux.dev, retrieved · first party docs · shows appearance, not capability
- Data
- /products/workmux.json · .md
- Method
- Same rules for all 103 researched ADEs. How we check · Report wrong data
What it does
Where you use it
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
Source and scope for Where you use itAgents it runs
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
Source and scope for Agents it runsAccounts and subscriptions
Bring your own agent subscriptions and logins: local runs reuse installed CLI logins; remote login executes each agent native login flow. No workmux account, credits, or local-model route; agent switching via named profiles.
Source and scope for Accounts and subscriptionsWhere agents run
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
Source and scope for Where agents runHow work is kept apart
Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed.
Source and scope for How work is kept apartParallel work
Independent agents run concurrently, each in its own worktree and multiplexer window/session; --max-concurrent bounds worker pools; running/waiting/failed/finished visible across dashboard, sidebar, list, and status.
Source and scope for Parallel workScheduled runs
Unattended triggers: the command reference pages plus additional CLI commands and the docs contain no schedules, event triggers, recurring tasks, or completion conditions; work starts from human or agent CLI invocations.
Source and scope for Scheduled runsGit and pull requests
Branch/worktree lifecycle with agent-driven commits, rebase command + skill with conflict intelligence, PR checkout/creation flow via gh, PR-check surfacing, and post-merge cleanup. No inline review comments or product-owned PR state.
Source and scope for Git and pull requestsLicence and source
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
Source and scope for Licence and sourcePrice
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
Source and scope for PriceTyped filters
Every value links to its receipt below. Read plan and surface restrictions before using a filter to decide.
| Filter | Value |
|---|---|
| Desktop OS: macOS | Yes cross-checked |
| Desktop OS: Windows | No cross-checked |
| Desktop OS: Linux | Yes cross-checked |
| Web client | No cross-checked |
| Phone access: Native iOS | No cross-checked |
| Phone access: Native Android | No cross-checked |
| Phone access: Phone browser | No cross-checked |
| CLI | Yes cross-checked |
| Coding agents: Claude Code | Yes cross-checked |
| Coding agents: Codex | Yes cross-checked |
| Coding agents: Cursor Agent | Unknown |
| Coding agents: Gemini CLI | Yes cross-checked |
| Coding agents: OpenCode | Yes cross-checked |
| Coding agents: Pi | Yes cross-checked |
| Coding agents: Grok | Yes cross-checked |
| Coding agents: Aider | Unknown |
| Coding agents: Antigravity | Yes cross-checked |
| Use my subscription | Yes cross-checked |
| Where agents run: On your computer | Yes cross-checked |
| Where agents run: On your own server | Yes cross-checked |
| Where agents run: In the vendor's cloud | No cross-checked |
| Workspace isolation: Git worktree | Yes cross-checked |
| Workspace isolation: Container | Yes cross-checked |
| Workspace isolation: VM or microVM | Yes cross-checked |
| Parallel runs | Yes cross-checked |
| Coordinator | Yes cross-checked |
| Native issue tracking | No cross-checked |
| Unattended triggers | No cross-checked |
| Shared prompting | No |
| Team roles | No |
| SSO | No cross-checked |
| Pull request workflow | Yes cross-checked |
| Review interface | Yes cross-checked |
| Product license: Apache-2.0 | No. MIT cross-checked |
| Product license: MIT | Yes. MIT cross-checked |
| Product license: GPL-3.0 | No. MIT cross-checked |
| Product license: GPL-3.0-or-later | No. MIT cross-checked |
| Product license: AGPL-3.0 | No. MIT cross-checked |
| Product license: AGPL-3.0-or-later | No. MIT cross-checked |
| Product license: Elastic License 2.0 | No. MIT cross-checked |
| Product license: Custom terms | No. MIT cross-checked |
| Product license: Proprietary | No. MIT cross-checked |
| Self-hosting | Yes cross-checked |
| Telemetry opt-out | No |
| Free access | Yes cross-checked |
| Starting price: Free | Yes. Free cross-checked |
| Starting price: Up to $20/month | No. Free cross-checked |
| Starting price: $21–40/month | No. Free cross-checked |
| Starting price: $41–100/month | No. Free cross-checked |
| Starting price: Over $100/month | No. Free cross-checked |
| Starting price: Contact for pricing | No. Free cross-checked |
| Pricing model: Free, no paid plan | Yes |
| Pricing model: Free plan | No |
| Pricing model: Subscription | No |
| Pricing model: Per seat | No |
| Pricing model: Usage-based | No |
| Pricing model: Per machine or compute | No |
| Pricing model: Bring your own key | Yes |
| Pricing model: One-time purchase | No |
| Paid plans from: No paid plan | Unknown |
| Paid plans from: $10 or less/month | Unknown |
| Paid plans from: $11–20/month | Unknown |
| Paid plans from: $21–40/month | Unknown |
| Paid plans from: $41–100/month | Unknown |
| Paid plans from: Over $100/month | Unknown |
| Paid plans from: Contact sales | Unknown |
| Lifecycle status: Active | Unknown |
| Lifecycle status: Sunsetting | Unknown |
| Keeps going with laptop closed | Yes cross-checked |
| Open-source license | Yes. MIT cross-checked |
| Any desktop app | Yes cross-checked |
| Paid plan up to $20/month | Unknown |
| Paid plan up to $40/month | Unknown |
| Any phone client | No cross-checked |
Filter receipts
Desktop OS: macOS: Yes cross-checked
Desktop OS: macOS
Typed value: Yes.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
Desktop OS: Windows: No cross-checked
Desktop OS: Windows
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Experimental kitty backend (remote-control API, Unix socket); Unix-like OS with named pipes required; Windows is not supported.
Desktop OS: Linux: Yes cross-checked
Desktop OS: Linux
Typed value: Yes.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
Web client: No cross-checked
Web client
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Cli: Documented command reference pages plus additional CLI-only commands (setup/send/wait), covering add/merge/status/list/dashboard/sidebar/remote/... on macOS/Linux.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
Phone access: Native iOS: No cross-checked
Phone access: Native iOS
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
Phone access: Native Android: No cross-checked
Phone access: Native Android
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
Phone access: Phone browser: No cross-checked
Phone access: Phone browser
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Cli: Documented command reference pages plus additional CLI-only commands (setup/send/wait), covering add/merge/status/list/dashboard/sidebar/remote/... on macOS/Linux.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
CLI: Yes cross-checked
CLI
Typed value: Yes.
Product CLI covers create/inspect/steer/stop/review with JSON output for status/list and agent-facing ops (status/send/wait/capture/run). No REST API, SDK, MCP server, or webhooks exist in the complete command inventory and docs.
- Cli: Documented command reference pages (24) plus additional CLI-only commands (setup/send/wait/capture/run), including add/merge/rebase/remove/list/status agent ops; shell completions included.
- Api: No REST or local HTTP API in the command inventory or full docs; machine access is CLI plus --json output.
- Sdk: No SDK in the command inventory or full docs; agents script the CLI instead.
- Mcp server: workmux exposes no MCP server; MCP appears only as agent-side config mounted into sandboxes by the underlying harness.
- Webhooks: No webhooks in the command inventory or full docs; triggers are human or agent CLI invocations.
- Operations: Create (add), inspect (status/list --json), steer (send/input), stop (dashboard kill/remove/reap-agents), review (diff/commit/merge) all exist as CLI or CLI-driven UI actions.
- No API/SDK/MCP-server/webhook surface: automation goes through the CLI.
- No published API-stability policy beyond versioned releases and remote protocol checks.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
Coding agents: Claude Code: Yes cross-checked
Coding agents: Claude Code
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: Codex: Yes cross-checked
Coding agents: Codex
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: Gemini CLI: Yes cross-checked
Coding agents: Gemini CLI
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: OpenCode: Yes cross-checked
Coding agents: OpenCode
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: Pi: Yes cross-checked
Coding agents: Pi
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: Grok: Yes cross-checked
Coding agents: Grok
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Coding agents: Antigravity: Yes cross-checked
Coding agents: Antigravity
Typed value: Yes.
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Harnesses: Built-in: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; status matrix confirms 9 with hook support, Kiro/Vibe referenced as agent commands.
- Custom adapters: Global named agents define type/command/args/env (incl. from_env), e.g. proxy and multi-account profiles; project configs reference but cannot define them.
- Session integration: Launches installed CLIs with native-format prompt injection; status hooks report working/waiting/done; --continue/--fork resume conversations; setup auto-detects installed agents.
- Mixed harness operation: Multi-generation (-a claude -a gemini) and layouts mixing agents per pane are documented; no mixed-harness run was observed in this pass.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
Use my subscription: Yes cross-checked
Use my subscription
Typed value: Yes.
Bring your own agent subscriptions and logins: local runs reuse installed CLI logins; remote login executes each agent native login flow. No workmux account, credits, or local-model route; agent switching via named profiles.
- Agent subscriptions: No workmux account or tokens: local agents run as installed CLIs under existing logins; remote login runs native flows (claude auth login, codex login, ...).
- Vendor account requirement: No workmux account, signup, or payment exists: install the binary and run. Agent-side logins are the only accounts involved.
- API-key and local-model routes were not established beyond proxy-style env examples.
- No vendor billing exists, so credits/overage behavior is not applicable.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
Where agents run: On your computer: Yes cross-checked
Where agents run: On your computer
Typed value: Yes.
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
- Local: Local worktrees with agent panes on the user machine; default placement.
- Remote backend is explicitly experimental.
- Survival covers the same laptop returning; continuing from another device is not documented.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
Where agents run: On your own server: Yes cross-checked
Where agents run: On your own server
Typed value: Yes.
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
- Byo remote: User-owned Linux servers over SSH (home box, Proxmox VM, cloud VM); user provisions, pays, and holds credentials.
- Managed cloud: No vendor cloud: placement is local or user remotes only; vendor operates no compute.
- Hybrid: Local and remote agents coexist in dashboard/sidebar/list/status; placement set per project or per add.
- Provisioning: remote setup installs node + hooks; server prerequisites (SSH, linger, toolchains) are user-provisioned with doctor checks.
- Payer: User pays: own hardware plus own servers; workmux itself charges nothing.
- Resources: Container CPU/memory limits configurable; server sizing and project toolchains are user-managed.
- Expiry: Remote workspaces persist (laptop may sleep); lifespan bounds undocumented; local reap-agents prunes old tracked processes.
- Client disconnect: Remote agents keep working through sleep, network loss, or closed windows; the local window reconnects by itself.
- Remote backend is explicitly experimental.
- Survival covers the same laptop returning; continuing from another device is not documented.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
Where agents run: In the vendor's cloud: No cross-checked
Where agents run: In the vendor's cloud
Typed value: No.
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
- Local: Local worktrees with agent panes on the user machine; default placement.
- Managed cloud: No vendor cloud: placement is local or user remotes only; vendor operates no compute.
- Hybrid: Local and remote agents coexist in dashboard/sidebar/list/status; placement set per project or per add.
- Remote backend is explicitly experimental.
- Survival covers the same laptop returning; continuing from another device is not documented.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
Workspace isolation: Git worktree: Yes cross-checked
Workspace isolation: Git worktree
Typed value: Yes.
Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed.
- Worktree: Git worktrees are the core isolation unit, each with its own window/panes.
- Process: Separate panes/processes per agent; sandboxed agents run in a separate OS instance; RPC bridge syncs host features.
- Documented threat model: agent edits are untrusted on rebuild; host git operations, hooks, and content filters sit outside the sandbox.
- Two-change code-separation validation was not performed in this pass.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
Workspace isolation: Container: Yes cross-checked
Workspace isolation: Container
Typed value: Yes.
Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed.
- Container: Docker, Podman, and Apple Container runtimes; ephemeral sessions; OCI runtime override (Kata) for hardware VM isolation.
- Vm: Lima VM backend for persistent machines with Nix/Devbox toolchains; documented reduced Git-metadata isolation acceptance.
- Os sandbox: Container/VM are the product sandboxes; agent-native OS sandboxes (Claude Seatbelt/bwrap, host-exec Seatbelt/bwrap) are harnesses, not workmux enforcement.
- Filesystem: Sandbox scoped to the worktree; host files and credentials stay outside; container guests cannot modify shared git config/hooks.
- Process: Separate panes/processes per agent; sandboxed agents run in a separate OS instance; RPC bridge syncs host features.
- Network: Sandbox network proxy with configurable limits and 503 backpressure; remote Docker-bridge firewalling advised; default-allow scope unclear.
- Secrets: Host SSH/AWS/credential files excluded; per-agent credential mounts; remote logins and env passthrough stored privately on the server.
- Security boundary: Documented boundary with explicit non-goals: untrusted edits, writable worktree programs, and host-side hooks are outside; "do not use" warning for strict threat models.
- Documented threat model: agent edits are untrusted on rebuild; host git operations, hooks, and content filters sit outside the sandbox.
- Two-change code-separation validation was not performed in this pass.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
Workspace isolation: VM or microVM: Yes cross-checked
Workspace isolation: VM or microVM
Typed value: Yes.
Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed.
- Container: Docker, Podman, and Apple Container runtimes; ephemeral sessions; OCI runtime override (Kata) for hardware VM isolation.
- Vm: Lima VM backend for persistent machines with Nix/Devbox toolchains; documented reduced Git-metadata isolation acceptance.
- Os sandbox: Container/VM are the product sandboxes; agent-native OS sandboxes (Claude Seatbelt/bwrap, host-exec Seatbelt/bwrap) are harnesses, not workmux enforcement.
- Filesystem: Sandbox scoped to the worktree; host files and credentials stay outside; container guests cannot modify shared git config/hooks.
- Process: Separate panes/processes per agent; sandboxed agents run in a separate OS instance; RPC bridge syncs host features.
- Network: Sandbox network proxy with configurable limits and 503 backpressure; remote Docker-bridge firewalling advised; default-allow scope unclear.
- Secrets: Host SSH/AWS/credential files excluded; per-agent credential mounts; remote logins and env passthrough stored privately on the server.
- Security boundary: Documented boundary with explicit non-goals: untrusted edits, writable worktree programs, and host-side hooks are outside; "do not use" warning for strict threat models.
- Documented threat model: agent edits are untrusted on rebuild; host git operations, hooks, and content filters sit outside the sandbox.
- Two-change code-separation validation was not performed in this pass.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
Parallel runs: Yes cross-checked
Parallel runs
Typed value: Yes.
Independent agents run concurrently, each in its own worktree and multiplexer window/session; --max-concurrent bounds worker pools; running/waiting/failed/finished visible across dashboard, sidebar, list, and status.
- Concurrent sessions: Each worktree gets its own window (or session in session mode); multi-generation creates many at once.
- Concurrent tasks: Parallel workflows: independent agents per worktree on different branches with isolated terminal state.
- Status visibility: Dashboard (priority-sorted), sidebar tiles, list table, and status --json all expose running/waiting/done/exited states.
- Concurrency is documented, not observed: no multi-agent run was executed in this pass.
- No documented CPU/RAM quotas; throttling is count-based only.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
Coordinator: Yes cross-checked
Coordinator
Typed value: Yes.
Product-shipped /coordinator and /worktree skills define a coordinator role that spawns worktree agents with prompt files, monitors via status/wait, sends follow-ups, reviews output, and merges. Mixed-harness and retained-relationship depth untested.
- Roles: Coordinator (plans/delegates, does not implement) vs worktree agents, defined by /coordinator and /worktree skills.
- Messages: Coordinator sends instructions via send/input/capture; -p/-P prompt injection at spawn.
- Results return: wait (with --timeout/--any and exit codes), capture output review, then merge; resumable across interruptions.
- Orchestration owner: Product-owned: skills ship in the repo and install via workmux setup; distinct from harness-native subagents (pi-suba noted as external publisher).
- Activation: Invoke /coordinator or /worktree after setup --skills; spawn with add -b -P from prompt files.
- Authentication: Local user context; sandboxed coordinators route through the host RPC bridge; no extra auth documented.
- Live progress: status table, wait, dashboard live preview, sidebar timers track delegated work.
- Retained results: Prompt files, captured output, and worktree contents persist until cleanup; no structured result store.
- Followup: Send follow-up instructions and re-wait; /merge after review; monitoring invariant survives interruptions.
- Invocation: Documented CLI (add -b -P/--name/--base) plus natural-language skill invocation.
- Delegation behavior is documented in the skill, not observed hands-on in this pass.
- Cross-harness teams and persistent relationship records were not established.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
Native issue tracking: No cross-checked
Native issue tracking
Typed value: No.
No issue objects exist: tracking is branch/worktree-based. Worktrees persist independently of windows, carry branch/age/status metadata, and need no external tracker; idea parking, discussion, and prioritization are absent.
- Create: No issue creation: work units are created as git branches/worktrees via add.
- Edit: No issue editing: changes happen in git and files directly.
- Discuss: No in-product discussion or comments surface in the inventory or docs.
- Label: No labels; branch names and project grouping are the only organizers.
- Assign: No assignment: single-user tool with no member model.
- Reopen: No reopen: resurrect restores windows, not issue states.
- Task session separation: Worktree (task unit) persists independently of windows/sessions: close keeps the worktree; session mode gives each worktree its own tmux session; resurrect restores windows after crashes.
- Workmux tracks worktrees/branches, not issues: no create/edit/discuss/prioritize/label/assign/search/filter/archive/reopen semantics.
- Findings rest on the CLI command inventory and docs; no hands-on fixture was run.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
Unattended triggers: No cross-checked
Unattended triggers
Typed value: No.
Unattended triggers: the command reference pages plus additional CLI commands and the docs contain no schedules, event triggers, recurring tasks, or completion conditions; work starts from human or agent CLI invocations.
- Schedules: No schedules: CLI command inventory has no schedule command; nothing in full docs.
- Events: No event triggers: Jira-style polling and git hooks are absent; pre_merge/pre_remove are lifecycle gates.
- Recurring tasks: No recurring tasks in inventory or docs.
- Scheduler location: No scheduler exists, so no location; closing the client ends nothing scheduled.
- Overlaps: No scheduler, so no overlap handling.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
Shared prompting: No
Shared prompting
Typed value: No.
Live human collaboration does not exist in the pinned release: a single-user local binary with no accounts, shared links, discussion, presence, or co-prompting; the dashboard `comment` action steers the operator own agent pane (diff_ops.rs). Bounded inspection of raine/workmux at e72fd59 for collaborat/multi-user/presence/co-prompt/share/sso found no collaboration surfaces.
- Visibility: No shared visibility: dashboard/sidebar/list are same-machine local views over local JSON state; no accounts or shared links.
- Discussion: No shared discussion: no issue threads or team chat surfaces.
- Comments: Dashboard `comment` (diff_ops.rs send_hunk_comment) pastes a hunk-annotated message into the operator own agent pane: single-user steering, not human comments.
- Presence: No presence: no users, sessions-by-person, or online indicators.
- Coprompting: No simultaneous co-prompting: one operator per pane; `send` targets the local pane.
- Ordering: No multi-user input ordering: no concurrent inputs to order.
- Conflicts: No conflict handling for co-prompting: none exists.
- Attribution: No attribution: no second actor exists to attribute.
- Permissions: No collaboration permissions: single OS user; sandbox mounts bound the harness, not people.
- Plan restrictions: No plans or tiers: the binary has no accounts, so no team or free-tier restrictions.
- Owner online: No collaboration session exists to require the owner: remote agents outlive laptop sleep for the same user, which is continuity, not co-presence.
-
Collaboration-surface inspection (dashboard comment action, local state store, remote model)
Finding: raine/workmux at e72fd59: the dashboard comment action (diff_ops.rs send_hunk_comment) pastes a hunk-annotated message into the operator own agent pane, which is single-user steering, not shared discussion; state is local per-pane JSON with no user, team, presence, or permission fields; remote servers are the single user SSH machines. Searched collaborat/multi-user/teammate/presence/co-prompt/share/comment/sso across src/, skills/, docs/: no shared visibility, discussion, presence, co-prompting, attribution, or plan-restriction surfaces.
Team roles: No
Team roles
Typed value: No.
Roles and access boundaries do not exist in the pinned release: no accounts, organizations, roles, membership, SSO, or provisioning; the product runs as the local OS user and remote credentials are the single user SSH keys and agent logins. Bounded inspection of raine/workmux at e72fd59 for sso/organization/role/acl/member found no role or access-boundary enforcement.
- Organization roles: No organizations: no accounts of any kind.
- Project roles: No project roles: grouping is by directory name, not membership.
- Task roles: No task roles: a single operator performs every action.
- Membership: No membership lifecycle: nothing to join or leave.
- Sso: No SSO: no authentication layer exists (local OS user; per-agent logins only).
- Provisioning: No provisioning: no users to provision; remote setup provisions machines, not accounts.
- Unauthorized access: No access-control model: single OS user; sandbox mounts restrict harness files, not human roles.
- Revocation: No membership to revoke: remote SSH keys are managed by the user outside the product.
- Plan scope: No plans: every control ships in the single free binary.
-
Access-model inspection (config schema, credentials, no accounts or roles)
Finding: raine/workmux at e72fd59: no accounts, organizations, roles, membership, SSO, or provisioning; the product runs as the local OS user with no signup (install guide); the config schema has no role or permission keys; remote/credentials.rs stores the single user SSH keys and per-agent logins. Searched sso/scim/organization/role/acl/member/policy across src/, skills/, docs/: no role or access-boundary enforcement beyond OS-user execution and sandbox file mounts.
SSO: No cross-checked
SSO
Typed value: No.
Approvals are agent-level (permission prompts, yolo opt-outs) with the sandbox as the product-enforced boundary; pre_merge hooks gate merges; login-time port forwarding is the only temporary access. No approver identities, lifetimes, or revocation model.
- Delivery: pre_merge hooks abort merges; dashboard diff review precedes merge; no enforced approval gate.
- Enforcement: Sandbox boundary and user hooks are enforced; prompt compliance is harness behavior.
- Product enforcement is the sandbox boundary plus user hooks; prompt-following behavior belongs to harnesses.
- No named approvers, scoped lifetimes, or revocation flow documented.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
Pull request workflow: Yes cross-checked
Pull request workflow
Typed value: Yes.
Branch/worktree lifecycle with agent-driven commits, rebase command + skill with conflict intelligence, PR checkout/creation flow via gh, PR-check surfacing, and post-merge cleanup. No inline review comments or product-owned PR state.
- Branches: Per-worktree branches with --base, fork refs, rename, and stacked --into merges.
- Pull requests: add --pr checks out PRs/MRs; /open-pr drafts and opens gh creation in browser without creating directly; remove --gone after merge.
- Commits and PR creation are agent/CLI-assisted, not product-native operations.
- Review comments and CI responses live in the hosting UI; workmux surfaces awareness only.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Template skill that drafts PR description from diff/commits and runs gh pr create --web to open creation in the browser; explicitly does not create the PR directly.
Review interface: Yes cross-checked
Review interface
Typed value: Yes.
Diff/patch review with file navigation, live terminal previews, PR links, and post-merge cleanup; prompt files and logs persist until removed. No artifact gallery, recordings, revision pinning, or leak analysis.
- Diffs: Dashboard diff view with syntax highlighting; list shows per-worktree diff stats; WIP vs committed toggle.
- Previews: Live terminal preview per agent; dev servers run as panes for manual preview; no hosted previews.
- Test outputs: Test runs visible in agent preview/output; no retained test-report artifacts.
- Task links: Worktree rows link branch, PR number/checks, and agent status; no task-registry links.
- Retention ends at cleanup: removal deletes the worktree and its review context.
- Reopening findings relies on harness history and git, not a product gallery.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
Product license: Apache-2.0: No. MIT cross-checked
Product license: Apache-2.0
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: MIT: Yes. MIT cross-checked
Product license: MIT
Typed value: Yes. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: GPL-3.0: No. MIT cross-checked
Product license: GPL-3.0
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: GPL-3.0-or-later: No. MIT cross-checked
Product license: GPL-3.0-or-later
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: AGPL-3.0: No. MIT cross-checked
Product license: AGPL-3.0
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: AGPL-3.0-or-later: No. MIT cross-checked
Product license: AGPL-3.0-or-later
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: Elastic License 2.0: No. MIT cross-checked
Product license: Elastic License 2.0
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: Custom terms: No. MIT cross-checked
Product license: Custom terms
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Product license: Proprietary: No. MIT cross-checked
Product license: Proprietary
Typed value: No. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Self-hosting: Yes cross-checked
Self-hosting
Typed value: Yes.
Self-contained local binary with no vendor service, accounts, or required backends; code is git-portable and uninstall removes data dirs. Product state files have no documented portability contract.
- Self hosting: Runs entirely on user machines/servers; no vendor service to host or depend on.
- Vendor dependencies: No required vendor services: install, config, git, and agents are all user-side; server needs no GitHub access.
- Cancellation: Uninstall script removes hooks, skills, cache, and state; Homebrew/Cargo paths documented; no account to close.
- Account independence: Fully usable with no vendor account; agent logins are harness-side only.
- State-file portability (agents dir, workmux.log) has no documented format contract.
- Agent-side history portability belongs to each harness, not the product.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
Telemetry opt-out: No
Telemetry opt-out
Typed value: No.
Code, prompts, credentials, and logs stay on user machines or user-owned servers; sandboxes exclude host secrets. No telemetry policy disclosed; only documented network behaviors are the daily update check (disableable) and release downloads.
- Telemetry controls: No product telemetry control or collection module in the inspected config/source inventory. The documented opt-out only disables update checks.
- No telemetry or data-handling policy page exists; assessment rests on architecture plus the two documented network behaviors.
- Model-provider data use belongs to each harness and its own policy.
-
workmux config and telemetry inspection
Finding: The pinned product has no telemetry configuration or telemetry implementation in the inspected source/dependency inventory. Its network opt-out auto_update_check/WORKMUX_NO_UPDATE_CHECK disables update checking, not product telemetry. Provider-owned agent traffic remains separate.
Free access: Yes cross-checked
Free access
Typed value: Yes.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: Free: Yes. Free cross-checked
Starting price: Free
Typed value: Yes. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: Up to $20/month: No. Free cross-checked
Starting price: Up to $20/month
Typed value: No. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: $21–40/month: No. Free cross-checked
Starting price: $21–40/month
Typed value: No. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: $41–100/month: No. Free cross-checked
Starting price: $41–100/month
Typed value: No. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: Over $100/month: No. Free cross-checked
Starting price: Over $100/month
Typed value: No. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Starting price: Contact for pricing: No. Free cross-checked
Starting price: Contact for pricing
Typed value: No. Free.
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Ongoing free: Free and open source with no paid tier; full binary free indefinitely.
- Trial: No trial: nothing to try out under time limit; the product is the free artifact.
- Credits trial: No credits trial: no credits exist.
- Self host cost: Self-run at $0 plus own hardware; open source with prebuilt binaries and source builds.
- Paid only: Not paid-only: every install path is free.
- Early access: Not gated: public repo, docs, and binaries with no waitlist.
- Card required: No card: no billing relationship exists.
- Paid evaluation: No payment or sales approval needed to evaluate; agent logins are the only external requirement.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Free, no paid plan: Yes
Pricing model: Free, no paid plan
Typed value: Yes.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Free plan: No
Pricing model: Free plan
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Subscription: No
Pricing model: Subscription
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Per seat: No
Pricing model: Per seat
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Usage-based: No
Pricing model: Usage-based
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Per machine or compute: No
Pricing model: Per machine or compute
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Pricing model: Bring your own key: Yes
Pricing model: Bring your own key
Typed value: Yes.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Model cost: Users supply coding-agent model credentials. The official sandbox setup explicitly supports passing the user Anthropic API key to the agent.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux provider key passthrough
Finding: The official agent-container setup explicitly passes the host ANTHROPIC_API_KEY into the selected coding-agent sandbox through the global env_passthrough setting. This is own-provider-key model access, separate from the free Workmux binary and any provider usage bill.
Pricing model: One-time purchase: No
Pricing model: One-time purchase
Typed value: No.
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- Currency: No currency: nothing is sold; install methods are all free.
- Interval: No billing interval: no subscription exists.
- Seats: No seats: single binary per user, no team billing.
- Seat minimum: No seat minimum: nothing is sold.
- Included usage: The full MIT binary is free. It bundles no vendor inference allowance; users supply coding-agent credentials and their own compute.
- Credits: No credits: no vendor metering exists.
- Trial: No trial: the full product is free, not time-boxed.
- Overages: No overages: no metered units.
- Taxes: No taxes: no sale occurs.
- Compute cost: Compute is BYO: own machine plus own servers; no vendor compute fees.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
Keeps going with laptop closed: Yes cross-checked
Keeps going with laptop closed
Typed value: Yes.
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
- Byo remote: User-owned Linux servers over SSH (home box, Proxmox VM, cloud VM); user provisions, pays, and holds credentials.
- Managed cloud: No vendor cloud: placement is local or user remotes only; vendor operates no compute.
- Hybrid: Local and remote agents coexist in dashboard/sidebar/list/status; placement set per project or per add.
- Expiry: Remote workspaces persist (laptop may sleep); lifespan bounds undocumented; local reap-agents prunes old tracked processes.
- Client disconnect: Remote agents keep working through sleep, network loss, or closed windows; the local window reconnects by itself.
- Remote backend is explicitly experimental.
- Survival covers the same laptop returning; continuing from another device is not documented.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
Open-source license: Yes. MIT cross-checked
Open-source license
Typed value: Yes. MIT.
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
- Client license: MIT: LICENSE grant plus Cargo.toml license field; the CLI/TUI binary is built from this source.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
Any desktop app: Yes cross-checked
Any desktop app
Typed value: Yes.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
Any phone client: No cross-checked
Any phone client
Typed value: No.
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Desktop os: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); Homebrew on macOS/Linux. Windows is explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
All public findings
Findings may summarize a source. They are not presented as verbatim statements.
workmux:ACC-01: Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported. cross-checked
workmux:ACC-01
Terminal-native access: macOS/Linux CLI plus tmux dashboard TUI and sidebar pane; kitty/WezTerm/Zellij backends experimental. No desktop GUI, mobile apps, browser client, or editor extensions. Windows explicitly unsupported.
- Only the cited atomic findings are established; no hands-on client matrix was run.
- Windows is explicitly unsupported; no mobile, browser, or editor-extension clients exist in the documented inventory.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
-
Finding: Experimental kitty backend (remote-control API, Unix socket); Unix-like OS with named pipes required; Windows is not supported.
workmux:ACC-02: Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here. cross-checked
workmux:ACC-02
Eleven built-in harnesses (claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok) launched as installed CLIs with prompt injection and status hooks; global named profiles allow custom commands/env; multi-harness layouts documented but not observed here.
- Model-provider choice is inherited from each harness; no provider catalog was established.
- Mixed-harness operation is documented (multi-agent generation, mixed layouts) but no two-harness run was observed in this pass.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
workmux:ACC-03: Bring your own agent subscriptions and logins: local runs reuse installed CLI logins; remote login executes each agent native login flow. No workmux account, credits, or local-model route; agent switching via named profiles. cross-checked
workmux:ACC-03
Bring your own agent subscriptions and logins: local runs reuse installed CLI logins; remote login executes each agent native login flow. No workmux account, credits, or local-model route; agent switching via named profiles.
- API-key and local-model routes were not established beyond proxy-style env examples.
- No vendor billing exists, so credits/overage behavior is not applicable.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Permission options: symlink .claude/settings.local.json across worktrees, or named agent shadowing claude with --dangerously-skip-permissions; separate work/personal configs via named agents or CLAUDE_CONFIG_DIR.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Experimental Zellij backend requiring 0.44.1+; operates in current session; window-mode only; no dashboard preview (dump-screen covers focused pane only); sidebar supported; Windows is not supported.
workmux:ACC-04: Install binary, optional init, then workmux add; works with local Git (local-state main detection) and hosted remotes (fetch refs, PR checkout). Requires an existing Git repo and tmux; no managed cloud exists; BYO-server onboarding via setup/doctor/login is experimental. cross-checked
workmux:ACC-04
Install binary, optional init, then workmux add; works with local Git (local-state main detection) and hosted remotes (fetch refs, PR checkout). Requires an existing Git repo and tmux; no managed cloud exists; BYO-server onboarding via setup/doctor/login is experimental.
- Remote-server onboarding is explicitly experimental.
- Behavior in a non-Git directory and managed-cloud setup (no such offering) bound the result; see atoms.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
workmux:ACC-05: Product CLI covers create/inspect/steer/stop/review with JSON output for status/list and agent-facing ops (status/send/wait/capture/run). No REST API, SDK, MCP server, or webhooks exist in the complete command inventory and docs. cross-checked
workmux:ACC-05
Product CLI covers create/inspect/steer/stop/review with JSON output for status/list and agent-facing ops (status/send/wait/capture/run). No REST API, SDK, MCP server, or webhooks exist in the complete command inventory and docs.
- No API/SDK/MCP-server/webhook surface: automation goes through the CLI.
- No published API-stability policy beyond versioned releases and remote protocol checks.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:ACC-06: Agents extend via six product-shipped skills and status plugins; integrations are product-managed through workmux setup with scoped credential mounts. workmux itself consumes no MCP and brokers no third-party OAuth. cross-checked
workmux:ACC-06
Agents extend via six product-shipped skills and status plugins; integrations are product-managed through workmux setup with scoped credential mounts. workmux itself consumes no MCP and brokers no third-party OAuth.
- MCP consumption belongs to the underlying harness; workmux only mounts agent MCP config into sandboxes.
- Permission scoping is via sandbox mounts, global-only agent definitions, and agent prompts rather than a product permission matrix.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:ACC-07: Rust CLI/TUI binary (Cargo, edition 2024) with prebuilt macOS/Linux binaries; no Electron/Tauri/native-GUI or mobile clients. cross-checked
workmux:ACC-07
Rust CLI/TUI binary (Cargo, edition 2024) with prebuilt macOS/Linux binaries; no Electron/Tauri/native-GUI or mobile clients.
- No desktop GUI or mobile clients exist; the client is the terminal binary plus tmux/sidebar/dashboard surfaces.
-
Finding: Rust binary crate name workmux, version 0.1.282, edition 2024, license MIT; homepage and repository point at github.com/raine/workmux.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
-
Finding: Experimental kitty backend (remote-control API, Unix socket); Unix-like OS with named pipes required; Windows is not supported.
-
Finding: Experimental Zellij backend requiring 0.44.1+; operates in current session; window-mode only; no dashboard preview (dump-screen covers focused pane only); sidebar supported; Windows is not supported.
workmux:PLAN-01: Local creation via add with prompt intake; PR/MR branch checkout (not issue import); git branches are the source of truth with branch-derived handles; PR awareness is read-only with remote cleanup. cross-checked
workmux:PLAN-01
Local creation via add with prompt intake; PR/MR branch checkout (not issue import); git branches are the source of truth with branch-derived handles; PR awareness is read-only with remote cleanup.
- No Linear/Jira-style issue import or two-way sync: external import is branch checkout only.
- No product-level plan editors; supported edits happen in git or the hosting UI.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Finding: Template skill that drafts PR description from diff/commits and runs gh pr create --web to open creation in the browser; explicitly does not create the PR directly.
workmux:PLAN-02: Durable task plans do not exist in the pinned release: prompts are one-shot launch briefs written to git-excluded `.workmux/PROMPT-<branch>.md` files (prompt.rs, prompt_loader.rs, setup.rs); the state store keeps pane-level status only (state/types.rs); dashboard/sidebar show agents and worktrees, not plans. Bounded inspection of raine/workmux at e72fd59 (homebrew-workmux tap holds packaging only) for plan/acceptance/revision/spec found no plan objects, editors, revision history, or implementation links. cross-checked
workmux:PLAN-02
Durable task plans do not exist in the pinned release: prompts are one-shot launch briefs written to git-excluded `.workmux/PROMPT-<branch>.md` files (prompt.rs, prompt_loader.rs, setup.rs); the state store keeps pane-level status only (state/types.rs); dashboard/sidebar show agents and worktrees, not plans. Bounded inspection of raine/workmux at e72fd59 (homebrew-workmux tap holds packaging only) for plan/acceptance/revision/spec found no plan objects, editors, revision history, or implementation links.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Prompt brief and state-store inspection (setup.rs, prompt.rs, state/types.rs)
Finding: Bounded inspection of raine/workmux at e72fd59 (homebrew-workmux tap holds packaging only, no product code): prompts are one-shot launch briefs written to git-excluded .workmux/PROMPT-<branch>.md (setup.rs write_prompt_file); prompt frontmatter parses only a foreach matrix (prompt.rs PromptMetadata); the state store keeps pane-level status/activity timestamps only (state/types.rs AgentState); dashboard/sidebar list agents and worktrees, not plans. Searched plan/acceptance/revision/spec across src/, skills/, docs/: no plan objects, acceptance criteria, editors, revision history, or implementation links.
workmux:PLAN-03: Hierarchy machinery does not exist in the pinned release: work is flat branches/worktrees; `merge --into` merges into any base branch (stacked git branches, not a product hierarchy); the state store has no parent/child/blocker/dependency fields; pre_merge/pre_remove are shell lifecycle gates; coordinator merge ordering is an agent prompt convention. Bounded inspection of raine/workmux at e72fd59 for epic/hierarchy/blocker/depend/readiness found no epics, blockers, readiness gates, dependency enforcement, or mandatory workflows. cross-checked
workmux:PLAN-03
Hierarchy machinery does not exist in the pinned release: work is flat branches/worktrees; `merge --into` merges into any base branch (stacked git branches, not a product hierarchy); the state store has no parent/child/blocker/dependency fields; pre_merge/pre_remove are shell lifecycle gates; coordinator merge ordering is an agent prompt convention. Bounded inspection of raine/workmux at e72fd59 for epic/hierarchy/blocker/depend/readiness found no epics, blockers, readiness gates, dependency enforcement, or mandatory workflows.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Hierarchy and dependency inspection (merge.rs, state/types.rs, config.rs, coordinator skill)
Finding: raine/workmux at e72fd59: merge --into targets any base branch (stacked git branches, not a product hierarchy); the state store is flat per-pane records with no parent/child/blocker/dependency fields; pre_merge/pre_remove are shell lifecycle gates, not dependency enforcement; coordinator merge ordering is an agent prompt convention (SKILL.md); multiplexer readiness is a shell-handshake wait (handshake.rs), not dispatch gating. Searched epic/hierarchy/blocker/depend/readiness/enforce across src/, skills/, docs/: no epics, blockers, readiness gates, dependency enforcement, state propagation, or mandatory workflows.
workmux:PLAN-04: Shared context and memory do not exist as a product surface in the pinned release: named-agent config and per-worktree PROMPT briefs are launch inputs, not a managed store; the config schema has no memory keys; the 6 bundled skills are static prompt files; sandbox and file seeding only mount or copy files. Bounded inspection of raine/workmux at e72fd59 for memory/instruction/provenance/retriev/index found no memory store and no scope, provenance, correction, deletion, or freshness controls. cross-checked
workmux:PLAN-04
Shared context and memory do not exist as a product surface in the pinned release: named-agent config and per-worktree PROMPT briefs are launch inputs, not a managed store; the config schema has no memory keys; the 6 bundled skills are static prompt files; sandbox and file seeding only mount or copy files. Bounded inspection of raine/workmux at e72fd59 for memory/instruction/provenance/retriev/index found no memory store and no scope, provenance, correction, deletion, or freshness controls.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Memory and context inspection (config.rs schema, skills.rs, sandbox mounts, file_ops.rs)
Finding: raine/workmux at e72fd59: the config schema holds named agents, remotes, hooks, layouts, and status icons with no memory/instruction/role keys (the only `memory` is Lima VM RAM); the 6 bundled skills are static prompt files installed to agent directories (skills.rs), not a managed store; sandbox mounts and file_ops only mount or seed credential, config, and prompt files. Searched memory/instruction/provenance/retriev/index across src/, skills/, docs/: no managed memory store and no scope, provenance, correction, deletion, or freshness controls.
workmux:PLAN-05: Architecture and API awareness do not exist as a product surface in the pinned release: the only built-in model call generates branch names (llm.rs); dashboard/sidebar filters match agent/worktree names; list columns carry git/status data only. Bounded inspection of raine/workmux at e72fd59 for search/index/retriev/citation/conflict/semantic found no code search, indexing, external-knowledge retrieval, citations, conflict detection, or review-time enforcement; those belong to the launched harnesses. cross-checked
workmux:PLAN-05
Architecture and API awareness do not exist as a product surface in the pinned release: the only built-in model call generates branch names (llm.rs); dashboard/sidebar filters match agent/worktree names; list columns carry git/status data only. Bounded inspection of raine/workmux at e72fd59 for search/index/retriev/citation/conflict/semantic found no code search, indexing, external-knowledge retrieval, citations, conflict detection, or review-time enforcement; those belong to the launched harnesses.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Retrieval and awareness inspection (llm.rs, dashboard/sidebar filters, list columns)
Finding: raine/workmux at e72fd59: the only built-in model call generates kebab-case git branch names (llm.rs generate_branch_name); dashboard/sidebar filters match agent/worktree names only; list columns are PROJECT/BRANCH/AGE/PR/AGENT/MUX/UNMERGED/PATH (git/status data). Searched search/index/retriev/citation/conflict/semantic/embed across src/, skills/, docs/: no product code search, repository indexing, external-knowledge retrieval, citations, conflict detection, or review-time enforcement; those belong to the launched harnesses.
workmux:PLAN-06: No issue objects exist: tracking is branch/worktree-based. Worktrees persist independently of windows, carry branch/age/status metadata, and need no external tracker; idea parking, discussion, and prioritization are absent. cross-checked
workmux:PLAN-06
No issue objects exist: tracking is branch/worktree-based. Worktrees persist independently of windows, carry branch/age/status metadata, and need no external tracker; idea parking, discussion, and prioritization are absent.
- Workmux tracks worktrees/branches, not issues: no create/edit/discuss/prioritize/label/assign/search/filter/archive/reopen semantics.
- Findings rest on the CLI command inventory and docs; no hands-on fixture was run.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
workmux:PLAN-07: Agents operate shared worktree handles through documented CLI ops (add/status/send/capture/wait/run) taught by product skills; human dashboard/sidebar/list views reflect agent-created work. No issue field model, claims, attribution, or conflict handling. cross-checked
workmux:PLAN-07
Agents operate shared worktree handles through documented CLI ops (add/status/send/capture/wait/run) taught by product skills; human dashboard/sidebar/list views reflect agent-created work. No issue field model, claims, attribution, or conflict handling.
- Handles are branch names, not issue records: no field edits, claims, assignment, or conflict semantics.
- Agent CLI surface (send/capture/wait/run) is documented in the skill, not the CLI reference; no hands-on verification here.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:PLAN-08: Living product specifications do not exist in the pinned release: PROMPT briefs are per-task launch files with no cross-task persistence, revisions, approvals, retrieval, or drift detection; the state store and dashboard hold no requirement objects. Bounded inspection of raine/workmux at e72fd59 for spec/requirement/persistent/approval/drift found no product-spec surface. cross-checked
workmux:PLAN-08
Living product specifications do not exist in the pinned release: PROMPT briefs are per-task launch files with no cross-task persistence, revisions, approvals, retrieval, or drift detection; the state store and dashboard hold no requirement objects. Bounded inspection of raine/workmux at e72fd59 for spec/requirement/persistent/approval/drift found no product-spec surface.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Spec persistence inspection (worktree skill, PROMPT brief lifecycle, state store)
Finding: raine/workmux at e72fd59: the /worktree skill dispatches by writing a per-task brief and running `workmux add`; the brief persists only as a git-excluded file in its worktree with no cross-task store, revisions, approvals, retrieval, or drift detection; the state store and dashboard hold no requirement objects. Searched spec/requirement/persistent/approval/drift across src/, skills/, docs/: no living product specification surface.
workmux:COORD-01: Independent agents run concurrently, each in its own worktree and multiplexer window/session; --max-concurrent bounds worker pools; running/waiting/failed/finished visible across dashboard, sidebar, list, and status. cross-checked
workmux:COORD-01
Independent agents run concurrently, each in its own worktree and multiplexer window/session; --max-concurrent bounds worker pools; running/waiting/failed/finished visible across dashboard, sidebar, list, and status.
- Concurrency is documented, not observed: no multi-agent run was executed in this pass.
- No documented CPU/RAM quotas; throttling is count-based only.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
workmux:COORD-02: Agents are placed per worktree with local/remote routing and worker-pool throttling; coordinator agents delegate via skills. No human assignment, claims, queues UI, or duplicate-work prevention. cross-checked
workmux:COORD-02
Agents are placed per worktree with local/remote routing and worker-pool throttling; coordinator agents delegate via skills. No human assignment, claims, queues UI, or duplicate-work prevention.
- Single-user tool: no human assignment or membership.
- Duplicate claims on the same task have no documented prevention; conflicts guidance is advisory.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Finding: New worktrees are clean checkouts needing files copy/symlink config; parallel branches can still conflict on merge; per-worktree installs recommended; node_modules cleanup handled in background.
workmux:COORD-03: Product-shipped /coordinator and /worktree skills define a coordinator role that spawns worktree agents with prompt files, monitors via status/wait, sends follow-ups, reviews output, and merges. Mixed-harness and retained-relationship depth untested. cross-checked
workmux:COORD-03
Product-shipped /coordinator and /worktree skills define a coordinator role that spawns worktree agents with prompt files, monitors via status/wait, sends follow-ups, reviews output, and merges. Mixed-harness and retained-relationship depth untested.
- Delegation behavior is documented in the skill, not observed hands-on in this pass.
- Cross-harness teams and persistent relationship records were not established.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:COORD-04: Unattended triggers: the command reference pages plus additional CLI commands and the docs contain no schedules, event triggers, recurring tasks, or completion conditions; work starts from human or agent CLI invocations. cross-checked
workmux:COORD-04
Unattended triggers: the command reference pages plus additional CLI commands and the docs contain no schedules, event triggers, recurring tasks, or completion conditions; work starts from human or agent CLI invocations.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:COORD-05: Stop (dashboard kill/remove/reap-agents), resume (resurrect/--continue/--fork/reconnect), steer (input mode/send), and waiting-state attention (priority sort, last-done) all documented; no pause or escalation chain. cross-checked
workmux:COORD-05
Stop (dashboard kill/remove/reap-agents), resume (resurrect/--continue/--fork/reconnect), steer (input mode/send), and waiting-state attention (priority sort, last-done) all documented; no pause or escalation chain.
- No pause primitive and no escalation chain documented.
- Error visibility rests on documented diagnostics (status --json, doctor); silent-wait behavior untested hands-on.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:COORD-06: Dashboard Agents/Worktrees views plus sidebar give a filterable fleet UI with attention sorting (waiting first), timers, diff stats, and navigation. No 12-task fixture was run; parent/child links and dependency views are absent. cross-checked
workmux:COORD-06
Dashboard Agents/Worktrees views plus sidebar give a filterable fleet UI with attention sorting (waiting first), timers, diff stats, and navigation. No 12-task fixture was run; parent/child links and dependency views are absent.
- No disclosed 12-state fixture run; live-agent verification absent in this pass.
- No parent/child tree or dependency rendering; coordinator relationships live in prompts/branches, not the UI.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
workmux:COORD-07: One-shot add -A and growing coordinator-led epics share the same tool; optional layouts, natural-language prompts, per-agent profiles, and mid-run steering are documented. No mandatory presets, in-run plan edits, or relationship records. cross-checked
workmux:COORD-07
One-shot add -A and growing coordinator-led epics share the same tool; optional layouts, natural-language prompts, per-agent profiles, and mid-run steering are documented. No mandatory presets, in-run plan edits, or relationship records.
- Growing-epic flow is coordinator-convention, not product-enforced hierarchy.
- Retained coordinator relationships and in-run plan edits were not established.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Always-visible agent overview pinned to tmux windows (or Zellij tabs): per-agent tile with status icon, worktree, elapsed time, project, git diff stats, optional PR number/checks, task description; toggle via workmux sidebar; session scoping and named views; requires status hooks.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:RUN-01: Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented. cross-checked
workmux:RUN-01
Local execution plus experimental BYO-server remotes over SSH in the same views; no vendor cloud or self-hosted service. Remote agents survive laptop sleep/disconnect with auto-reconnect; other-device continuation undocumented.
- Remote backend is explicitly experimental.
- Survival covers the same laptop returning; continuing from another device is not documented.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
workmux:RUN-02: Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed. cross-checked
workmux:RUN-02
Git worktree/branch file isolation plus container (Docker/Podman/Apple) and VM (Lima/Kata) sandboxes with scoped mounts and secret exclusion. Host-side git ops stay outside the boundary; hostile-code containment is explicitly disclaimed.
- Documented threat model: agent edits are untrusted on rebuild; host git operations, hooks, and content filters sit outside the sandbox.
- Two-change code-separation validation was not performed in this pass.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: New worktrees are clean checkouts needing files copy/symlink config; parallel branches can still conflict on merge; per-worktree installs recommended; node_modules cleanup handled in background.
workmux:RUN-03: Crash recovery via resurrect, resumable remote provisioning, automatic remote reconnect, and state-file reconciliation with diagnostics; transcript resume via --continue/--fork. Multi-day continuity and loss/duplication behavior untested. cross-checked
workmux:RUN-03
Crash recovery via resurrect, resumable remote provisioning, automatic remote reconnect, and state-file reconciliation with diagnostics; transcript resume via --continue/--fork. Multi-day continuity and loss/duplication behavior untested.
- Recovery is documented, not exercised: no crash/disconnect drill was run in this pass.
- Context loss, duplicate actions, and multi-day continuity were not established.
-
Finding: Permission options: symlink .claude/settings.local.json across worktrees, or named agent shadowing claude with --dangerously-skip-permissions; separate work/personal configs via named agents or CLAUDE_CONFIG_DIR.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
workmux:RUN-04: Dependencies via hooks and pane installs, secrets via files copy, caches via symlinks, dev servers as panes, remote node checks; per-worktree installs with background cleanup. No snapshots or managed services. cross-checked
workmux:RUN-04
Dependencies via hooks and pane installs, secrets via files copy, caches via symlinks, dev servers as panes, remote node checks; per-worktree installs with background cleanup. No snapshots or managed services.
- Environment repair is operator-driven (doctor fixes, retry/resume); app-level failures are the user/agent job.
- Service orchestration is panes-plus-docs (direnv port guidance), not managed services.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: New worktrees are clean checkouts needing files copy/symlink config; parallel branches can still conflict on merge; per-worktree installs recommended; node_modules cleanup handled in background.
workmux:RUN-05: Count-based throttling (--max-concurrent), wait timeouts, process-age reaping, pre-merge abort, and sandbox proxy limits exist; no time/turn/token/spend budgets or provider-failure policy documented. cross-checked
workmux:RUN-05
Count-based throttling (--max-concurrent), wait timeouts, process-age reaping, pre-merge abort, and sandbox proxy limits exist; no time/turn/token/spend budgets or provider-failure policy documented.
- No budgets for time, turns, tokens, or spend; no provider-failure retry policy.
- Enforcement is local abort/exit-code semantics, not billing controls.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:RUN-06: Host log, reconciled state files with --json diagnostics, per-agent/row stats (status, elapsed, titles, diffs), waiting/done failure semantics, and doctor checks. No traces, usage exports, aggregates beyond views, or measured performance claims. cross-checked
workmux:RUN-06
Host log, reconciled state files with --json diagnostics, per-agent/row stats (status, elapsed, titles, diffs), waiting/done failure semantics, and doctor checks. No traces, usage exports, aggregates beyond views, or measured performance claims.
- Truthfulness rests on documentation: no induced failure or reconnect-consistency drill was run here.
- No traces, cost/usage exports, or disclosed speed benchmarks.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: remote lists servers and placement default; setup installs matching node + status hooks; doctor runs 10 checks (ssh/arch/node/git/tmux/linger/mode/hooks/logins/image) with fixes; login runs native agent flows with --import/--setup-token options.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:RUN-07: Per-task API and compute cost visibility does not exist in the pinned release: list, status, and dashboard expose git and agent-status data only; repo-wide inspection finds no spend, token-usage, allowance, compute-fee, alert, export, or drill-down code (`token` means auth tokens, `budget` means UI widths and a transfer cap). Bounded inspection of raine/workmux at e72fd59 (src/, skills/, docs/) found no cost surface. cross-checked
workmux:RUN-07
Per-task API and compute cost visibility does not exist in the pinned release: list, status, and dashboard expose git and agent-status data only; repo-wide inspection finds no spend, token-usage, allowance, compute-fee, alert, export, or drill-down code (`token` means auth tokens, `budget` means UI widths and a transfer cap). Bounded inspection of raine/workmux at e72fd59 (src/, skills/, docs/) found no cost surface.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Cost-visibility inspection (list.rs columns, status output, dashboard; cost/spend grep)
Finding: raine/workmux at e72fd59: list/status/dashboard expose git and agent-status data only (PROJECT/BRANCH/AGE/PR/AGENT/MUX/UNMERGED/PATH); repo-wide grep for cost/spend/budget/billing/invoice finds only UI width budgets, a 64 MiB transfer byte budget, and claude setup-token auth flows, never usage. No model/API spend, token usage, subscription allowance, compute fees, alerts, export, or drill-down anywhere in src/, skills/, or docs/.
workmux:RUN-08: Count-based concurrency (--max-concurrent), wait-any wakeups, container CPU/RAM caps, proxy backpressure, and kill/sweep cancellation with resurrect/remote recovery. No host-aware scheduling, check reuse, fairness, or shared-limit model. cross-checked
workmux:RUN-08
Count-based concurrency (--max-concurrent), wait-any wakeups, container CPU/RAM caps, proxy backpressure, and kill/sweep cancellation with resurrect/remote recovery. No host-aware scheduling, check reuse, fairness, or shared-limit model.
- Scheduling is throttling plus operator cancellation, not a resource-aware scheduler.
- No bounded multi-check load test exists or was run.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:TEAM-01: Live human collaboration does not exist in the pinned release: a single-user local binary with no accounts, shared links, discussion, presence, or co-prompting; the dashboard `comment` action steers the operator own agent pane (diff_ops.rs). Bounded inspection of raine/workmux at e72fd59 for collaborat/multi-user/presence/co-prompt/share/sso found no collaboration surfaces. cross-checked
workmux:TEAM-01
Live human collaboration does not exist in the pinned release: a single-user local binary with no accounts, shared links, discussion, presence, or co-prompting; the dashboard `comment` action steers the operator own agent pane (diff_ops.rs). Bounded inspection of raine/workmux at e72fd59 for collaborat/multi-user/presence/co-prompt/share/sso found no collaboration surfaces.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Collaboration-surface inspection (dashboard comment action, local state store, remote model)
Finding: raine/workmux at e72fd59: the dashboard comment action (diff_ops.rs send_hunk_comment) pastes a hunk-annotated message into the operator own agent pane, which is single-user steering, not shared discussion; state is local per-pane JSON with no user, team, presence, or permission fields; remote servers are the single user SSH machines. Searched collaborat/multi-user/teammate/presence/co-prompt/share/comment/sso across src/, skills/, docs/: no shared visibility, discussion, presence, co-prompting, attribution, or plan-restriction surfaces.
workmux:TEAM-02: Roles and access boundaries do not exist in the pinned release: no accounts, organizations, roles, membership, SSO, or provisioning; the product runs as the local OS user and remote credentials are the single user SSH keys and agent logins. Bounded inspection of raine/workmux at e72fd59 for sso/organization/role/acl/member found no role or access-boundary enforcement. cross-checked
workmux:TEAM-02
Roles and access boundaries do not exist in the pinned release: no accounts, organizations, roles, membership, SSO, or provisioning; the product runs as the local OS user and remote credentials are the single user SSH keys and agent logins. Bounded inspection of raine/workmux at e72fd59 for sso/organization/role/acl/member found no role or access-boundary enforcement.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Access-model inspection (config schema, credentials, no accounts or roles)
Finding: raine/workmux at e72fd59: no accounts, organizations, roles, membership, SSO, or provisioning; the product runs as the local OS user with no signup (install guide); the config schema has no role or permission keys; remote/credentials.rs stores the single user SSH keys and per-agent logins. Searched sso/scim/organization/role/acl/member/policy across src/, skills/, docs/: no role or access-boundary enforcement beyond OS-user execution and sandbox file mounts.
workmux:TEAM-03: Approvals are agent-level (permission prompts, yolo opt-outs) with the sandbox as the product-enforced boundary; pre_merge hooks gate merges; login-time port forwarding is the only temporary access. No approver identities, lifetimes, or revocation model. cross-checked
workmux:TEAM-03
Approvals are agent-level (permission prompts, yolo opt-outs) with the sandbox as the product-enforced boundary; pre_merge hooks gate merges; login-time port forwarding is the only temporary access. No approver identities, lifetimes, or revocation model.
- Product enforcement is the sandbox boundary plus user hooks; prompt-following behavior belongs to harnesses.
- No named approvers, scoped lifetimes, or revocation flow documented.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Permission options: symlink .claude/settings.local.json across worktrees, or named agent shadowing claude with --dangerously-skip-permissions; separate work/personal configs via named agents or CLAUDE_CONFIG_DIR.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
workmux:TEAM-04: History lives in git (commits, branches) plus worktree metadata, dashboard diffs, and machine-readable status JSON. No decision records, comments, cross-user persistence, export feature, or integrity model. cross-checked
workmux:TEAM-04
History lives in git (commits, branches) plus worktree metadata, dashboard diffs, and machine-readable status JSON. No decision records, comments, cross-user persistence, export feature, or integrity model.
- No audit vocabulary is warranted: no retention, integrity, or tamper model documented.
- Machine-readable outputs exist, but no history export feature.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
workmux:TEAM-05: Dashboard-centered review: waiting-first surfacing, diff/patch evidence with syntax highlighting, PR check links, keyboard-first navigation with mouse option, and commit/merge/sweep actions. No comments, effort data, or guided next-action engine. cross-checked
workmux:TEAM-05
Dashboard-centered review: waiting-first surfacing, diff/patch evidence with syntax highlighting, PR check links, keyboard-first navigation with mouse option, and commit/merge/sweep actions. No comments, effort data, or guided next-action engine.
- Findings rest on documentation; the shared review fixture was not executed here.
- No commenting, change-request flow, or high-volume triage beyond sort/filter/sweep.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:TEAM-06: Colleague takeover does not exist in the pinned release: a single-user tool with no second-user, ownership-transfer, or attribution surfaces; same-user resume (--continue/--fork, resurrect) is explicitly not full takeover. Bounded inspection of raine/workmux at e72fd59 for takeover/handoff/colleague/ownership found no handover surfaces; artifacts move only through git and the filesystem. cross-checked
workmux:TEAM-06
Colleague takeover does not exist in the pinned release: a single-user tool with no second-user, ownership-transfer, or attribution surfaces; same-user resume (--continue/--fork, resurrect) is explicitly not full takeover. Bounded inspection of raine/workmux at e72fd59 for takeover/handoff/colleague/ownership found no handover surfaces; artifacts move only through git and the filesystem.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
-
Takeover and resume inspection (conversation forkers, resurrect, no multi-user)
Finding: raine/workmux at e72fd59: --continue/--fork resume the same user same-harness conversation from that harness own store (per-harness ConversationForker implementations); switching harness requires a new worktree via -a; resurrect restores windows after crashes for the same user and machine. Searched takeover/handoff/colleague/second-user/ownership/attribution across src/, skills/, docs/: no second-user, ownership-transfer, or attribution surfaces; artifacts (branches, files) move only through git and the filesystem, not product takeover.
workmux:SHIP-01: Diff/patch review with file navigation, live terminal previews, PR links, and post-merge cleanup; prompt files and logs persist until removed. No artifact gallery, recordings, revision pinning, or leak analysis. cross-checked
workmux:SHIP-01
Diff/patch review with file navigation, live terminal previews, PR links, and post-merge cleanup; prompt files and logs persist until removed. No artifact gallery, recordings, revision pinning, or leak analysis.
- Retention ends at cleanup: removal deletes the worktree and its review context.
- Reopening findings relies on harness history and git, not a product gallery.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
workmux:SHIP-02: User-configured hooks and panes run tests/lint/types/CI with pre_merge abort; coordinator and human diff review act as independent review. No built-in gates, required checks, or spec conformance. cross-checked
workmux:SHIP-02
User-configured hooks and panes run tests/lint/types/CI with pre_merge abort; coordinator and human diff review act as independent review. No built-in gates, required checks, or spec conformance.
- All verification is user-configured; the product ships no default required checks.
- An agent saying checks passed is not a gate: only hook exit codes and human review enforce.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
workmux:SHIP-03: Branch/worktree lifecycle with agent-driven commits, rebase command + skill with conflict intelligence, PR checkout/creation flow via gh, PR-check surfacing, and post-merge cleanup. No inline review comments or product-owned PR state. cross-checked
workmux:SHIP-03
Branch/worktree lifecycle with agent-driven commits, rebase command + skill with conflict intelligence, PR checkout/creation flow via gh, PR-check surfacing, and post-merge cleanup. No inline review comments or product-owned PR state.
- Commits and PR creation are agent/CLI-assisted, not product-native operations.
- Review comments and CI responses live in the hosting UI; workmux surfaces awareness only.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
-
Finding: Template skill that drafts PR description from diff/commits and runs gh pr create --web to open creation in the browser; explicitly does not create the PR directly.
-
Finding: New worktrees are clean checkouts needing files copy/symlink config; parallel branches can still conflict on merge; per-worktree installs recommended; node_modules cleanup handled in background.
workmux:SHIP-04: The local user approves and merges (merge/rebase/squash strategies, --into targets, keep controls); pre_merge hooks gate merges. No release pipeline, deployment authority, or multi-party approval. cross-checked
workmux:SHIP-04
The local user approves and merges (merge/rebase/squash strategies, --into targets, keep controls); pre_merge hooks gate merges. No release pipeline, deployment authority, or multi-party approval.
- Single-user authority: no approver roles or deployment targets.
- Exact-revision guarantees are git semantics; no product revision pinning.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
workmux:SHIP-05: Branch-to-PR links with check status, merge-then-cleanup delivery states, sweep staleness reasons, and git/log/prompt evidence until cleanup. No rollback/revert commands and no stale-approval model. cross-checked
workmux:SHIP-05
Branch-to-PR links with check status, merge-then-cleanup delivery states, sweep staleness reasons, and git/log/prompt evidence until cleanup. No rollback/revert commands and no stale-approval model.
- Undo is manual git (revert/reflog), not a product operation.
- Delivery status ends at merge/cleanup; deployments are out of scope.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Command reference inventory: 24 documented command pages (add merge rebase remove rename list status open close resurrect sync-files path shell dashboard sidebar reap-agents config edit init claude prune sandbox remote completions docs update last-done) plus additional CLI-only commands (setup, send, wait, capture, run); no REST API, SDK, MCP server, webhook, or scheduling commands appear in the reference or full docs.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
workmux:SHIP-06: Dashboard Agents/Worktrees tables form a review queue (sort/filter/jump, diff, input-steer, commit/merge keys) with sweep bulk-removal of merged/gone work. No summaries, risk scoring, revision approvals, merge queues, or dedup. cross-checked
workmux:SHIP-06
Dashboard Agents/Worktrees tables form a review queue (sort/filter/jump, diff, input-steer, commit/merge keys) with sweep bulk-removal of merged/gone work. No summaries, risk scoring, revision approvals, merge queues, or dedup.
- Batch review is per-agent triage, not a coordinated release queue: no dependency ordering or approval pinning.
- The six-plus-item fixture review was not executed hands-on here.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Coordinator agent uses workmux add/status/send/capture/wait/run; write-prompts-first-then-spawn; workmux wait with --timeout/--any and exit codes 0/1/2/3; continuous-monitoring invariant across interruptions; cross-project project:handle targeting.
workmux:OWN-01: MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions. cross-checked
workmux:OWN-01
MIT-licensed Rust binary and repository (LICENSE + Cargo.toml at pinned v0.1.282); public source on GitHub; no component or hosted-service exceptions.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Rust binary crate name workmux, version 0.1.282, edition 2024, license MIT; homepage and repository point at github.com/raine/workmux.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: Tag v0.1.282 published 2026-11-10; most recent release observed 2026-11-12.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
workmux:OWN-02: Self-contained local binary with no vendor service, accounts, or required backends; code is git-portable and uninstall removes data dirs. Product state files have no documented portability contract. cross-checked
workmux:OWN-02
Self-contained local binary with no vendor service, accounts, or required backends; code is git-portable and uninstall removes data dirs. Product state files have no documented portability contract.
- State-file portability (agents dir, workmux.log) has no documented format contract.
- Agent-side history portability belongs to each harness, not the product.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Creates worktree + multiplexer window; --pr checks out GitHub PRs (gh) or GitLab MRs (glab 1.37+, --forge); -A auto-name via llm CLI; --fork/--continue resume; -n/--foreach/--agent multi-generation; -l layouts; --remote/--local placement; -b background; --with-changes moves uncommitted work.
-
Finding: Local git merge into main (or --into) with merge/rebase/squash strategies plus worktree+window+branch cleanup; --keep/merge_keep; --notification system notification; PR workflows merge remotely and use remove for cleanup instead.
-
Finding: Lists worktrees with AGE, AGENT status icons (incl. counts, text labels when piped), MUX window presence, UNMERGED marker; --pr shows PR status (gh); --json output; remote workspaces listed with remote name, last-known state, and exited reasons such as host reboot.
-
Finding: Reports tracked agents per repository (or --all/project:handle); --json emits observation with backend context, state-file reconciliation counts, per-agent worktree/branch/status/elapsed/title/pane/git, and target_errors with stable codes.
workmux:OWN-03: Code, prompts, credentials, and logs stay on user machines or user-owned servers; sandboxes exclude host secrets. No telemetry policy disclosed; only documented network behaviors are the daily update check (disableable) and release downloads.
workmux:OWN-03
Code, prompts, credentials, and logs stay on user machines or user-owned servers; sandboxes exclude host secrets. No telemetry policy disclosed; only documented network behaviors are the daily update check (disableable) and release downloads.
- No telemetry or data-handling policy page exists; assessment rests on architecture plus the two documented network behaviors.
- Model-provider data use belongs to each harness and its own policy.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Supported: Claude Code, OpenCode, Codex, Copilot CLI, Pi, Oh My Pi, Gemini CLI, Grok, Antigravity agy (with version/capability footnotes); Kiro and Mistral Vibe are tracking-issues only; icons working/waiting/done; workmux setup installs hooks; last-done/last-agent navigation; nested-agent opt-out via WORKMUX_DISABLE_SET_WINDOW_STATUS=1.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Container (Docker/Podman/Apple Container) and Lima VM backends isolate agents from host secrets; sandboxed agents get noninteractive prompts with sandbox as boundary; status/dashboard/merge work the same via RPC bridge; documented limits: edits untrusted, host git ops and hooks outside boundary, no hostile-code-containment promise.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: New worktrees are clean checkouts needing files copy/symlink config; parallel branches can still conflict on merge; per-worktree installs recommended; node_modules cleanup handled in background.
-
Finding: Prebuilt binaries for macOS (Apple Silicon/Intel) and Linux (x86_64/ARM64); background update check at most daily, disableable via auto_update_check:false or WORKMUX_NO_UPDATE_CHECK=1; update verifies checksums, Homebrew installs defer to brew.
-
workmux config and telemetry inspection
Finding: The pinned product has no telemetry configuration or telemetry implementation in the inspected source/dependency inventory. Its network opt-out auto_update_check/WORKMUX_NO_UPDATE_CHECK disables update checking, not product telemetry. Provider-owned agent traffic remains separate.
workmux:OWN-04: Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
workmux:OWN-04
Workmux is a free MIT-licensed application with no paid Workmux tier. Model use runs through the selected coding agent credentials; sandbox setup explicitly supports the user Anthropic API key. Provider usage and user compute are separate costs.
- The free application includes no hosted model allowance. Agent/model usage and user-operated compute have separate costs and limits.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Global ~/.config/workmux/config.yaml plus per-project .workmux.yaml; remotes and named agents are global-only; placement selects local or a named remote; panes support <agent> placeholders, splits, sizes; files copy/symlink; post_create/pre_merge/pre_remove hooks.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
workmux provider key passthrough
Finding: The official agent-container setup explicitly passes the host ANTHROPIC_API_KEY into the selected coding-agent sandbox through the global env_passthrough setting. This is own-provider-key model access, separate from the free Workmux binary and any provider usage bill.
workmux:OWN-05: Platform cost is $0 for one, two, or five developers (each installs the free binary on own hardware). Variable costs are agent subscriptions/API usage and any user-owned server spend, both unpriced in product sources. cross-checked
workmux:OWN-05
Platform cost is $0 for one, two, or five developers (each installs the free binary on own hardware). Variable costs are agent subscriptions/API usage and any user-owned server spend, both unpriced in product sources.
- Variable costs (agent plans, API meters, server bills) are unknown from product sources and excluded from the $0 figure.
- No per-task token/time assumptions exist to price; the catalog forbids inventing task costs.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
workmux:OWN-06: Actively maintained 0.x (v0.1.282 on 2026-11-10, weekly cadence) with public docs, changelog, and GitHub-issue support; experimental subsystems flagged. No SLA, status page, or access restrictions. cross-checked
workmux:OWN-06
Actively maintained 0.x (v0.1.282 on 2026-11-10, weekly cadence) with public docs, changelog, and GitHub-issue support; experimental subsystems flagged. No SLA, status page, or access restrictions.
- 0.x versioning with explicitly experimental backends (remote, kitty/WezTerm/Zellij); availability beta reflects that, not downtime.
- No support commitments, status history, or deprecation policy documented.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
raine/workmux repository metadata (GitHub API)
Finding: 3037 stars, 314 forks, 64 open issues observed 2026-11-12; pushed 2026-11-10; MIT license; owner login raine.
-
Finding: Manages git worktrees + tmux windows for parallel AI agents; features dashboard, sidebar, /worktree skill, agent status tracking, sandbox, remote servers, branch-name generation; install via script/Homebrew/Cargo/mise/Nix; philosophy "tmux is the interface".
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Six product-shipped skills; workmux setup --skills installs to Claude/OpenCode/Pi/OMP dirs (manual copy documented); dashboard m keybinding triggers /merge; /merge commits/rebases/merges with conflict review; /worktree delegates with --fork option; /coordinator manages full lifecycle.
-
Finding: add -A auto-names from prompt; /worktree spins off from running agents; --fork resumes conversations (Claude/Codex/Pi); /coordinator plans+delegates+merges; /merge for direct merge; /open-pr + git push + gh pr create then workmux remove/--gone for PR flow.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
-
Finding: Tag v0.1.282 published 2026-11-10; most recent release observed 2026-11-12.
workmux:OWN-08: Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios. cross-checked
workmux:OWN-08
Ongoing free MIT download with full functionality and no card, trial, or paid evaluation path; agent and compute costs stay separate. Single-user scope bounds team scenarios.
- Two-person collaboration scenario: each person free, but no shared/multiplayer surface exists (see TEAM-01).
- Meaningful evaluation needs a coding-agent login (BYO subscription), not product payment.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: MIT License grant; copyright (c) 2025 workmux contributors; no component exceptions in the file.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Prerequisites: Git 2.5+ and tmux (or WezTerm/kitty/Zellij); flow is install, optional workmux init, workmux add (creates worktree + window + panes + setup), work, workmux merge cleanup; /worktree skill recommended from inside agent sessions.
-
Finding: Optional TUI popup (tmux display-popup binding) requiring status hooks; Agents view (status, git info, PR checks, live preview) and Worktrees view; input mode sends keystrokes to agent; keys jump/diff/commit/merge/kill/remove/sweep; sort by priority/project/recency; sweep bulk-removes merged/gone worktrees.
-
Finding: Experimental: run agents on user-reachable Linux servers over SSH; each remote agent keeps a local tmux window and appears in dashboard/sidebar/status/list as handle@remote; agent keeps working while local machine sleeps, loses network, or closes the window; window reconnects by itself; workmux merge brings work home; local machine stays Git source of truth.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.
workmux:OWN-09: Transparent: the ADE is free with no paid units, limits, overages, markups, minimums, or cancellable plans; agent/compute spend is third-party and out of scope. cross-checked
workmux:OWN-09
Transparent: the ADE is free with no paid units, limits, overages, markups, minimums, or cancellable plans; agent/compute spend is third-party and out of scope.
-
workmux - parallel AI agents in your terminal
Finding: Terminal-first workflow for parallel AI agents using git worktrees: one task per worktree and tmux window, sidebar and dashboard for agent status, sandbox containment, SSH remote servers, agent skills; described as free and open source.
-
raine/workmux (GitHub repository)
Finding: Owner raine (Raine Virta); description "git worktrees + tmux windows for zero-friction parallel dev"; homepage https://workmux.dev/; MIT license; default branch main; not archived.
-
Finding: Install via install script, Homebrew (macOS/Linux), Cargo, mise, Nix, or prebuilt binaries; shell completions for bash/zsh/fish; uninstall script removes data dirs; no account, signup, or payment step.
-
Finding: Built-in agents: claude, codex, copilot, opencode, gemini, agy, kiro-cli, vibe, pi, omp, grok; prompt injection via -p/-P/-e without config changes; named global profiles with type/command/args/env (incl. from_env); multi-worktree generation (-a/-n/--foreach) e.g. one worktree each for claude and gemini.
-
Finding: Requires Linux x86-64/ARM64, promptless SSH, git, tmux 3.3+, systemd linger, API access; host mode (no sandbox) vs Docker mode per sandbox.enabled/isolation; server never talks to Git hosting and gets no origin; remote_files allowlist (64 MiB cap); no host-command proxying on servers.
-
Finding: remote login runs each agent native login flow (claude/codex/gemini/opencode/pi/omp) with localhost port forwarding; --import does a one-time credential copy (with rotation caveats); setup-token for Claude; sandbox env passthrough read once per workspace and stored privately on server.
-
Finding: Latest v0.1.282 (2026-11-10); near-weekly releases through 2026; recent additions: remote servers (v0.1.278), branch auto-name, coordinator commands, Zellij sidebar, Copilot/Grok/agy support; entries describe shipped releases only, no future-dated commitments found.