# Claude Code

Podium publishes this catalog and sells an agent development environment. Podium chooses the products, criteria, sources and wording, so treat these pages as vendor-authored research. Check important claims against the linked primary sources. The same evidence rules apply to Podium and its competitors, but they do not remove this conflict of interest.

Page: https://adecomparison.com/ade/claude-code/
Website: https://code.claude.com/docs/en/overview
Research: covered

## Measured by ADEMark

Source: https://ademark.dev/ (CC BY 4.0; credit ADEMark). Claude Code 4.23118.0, measured up to 12 November 2026 on Mac mini M4, 16 GB, macOS 27 · Claude Opus 5.5, effort high. Third-party measurements, not findings; no overall score. Place counts only apps with a value.

- Cold start: 2.43 s (7th of 12), median of 9 runs
- New thread, ready to type: Not measured. ADEMark: it talks only to its maker's cloud, so our local replay server cannot answer it
- First word on screen: Not measured. ADEMark: it talks only to its maker's cloud, so our local replay server cannot answer it
- Typing lag, 5 agents streaming: Not measured. ADEMark: it talks only to its maker's cloud, so our local replay server cannot answer it
- Open a 2,000-message thread: Not measured. ADEMark: it talks only to its maker's cloud, so our local replay server cannot answer it
- Memory at rest, 5 threads: 1.66 GB (10th of 12), 1 run
- Idle CPU: 0.71% (4th of 13), 1 run
- Memory under load: Not measured. ADEMark: it talks only to its maker's cloud, so our local replay server cannot answer it
- Tokens to answer “hi”: 75.5K tokens (12th of 13), median of 3 runs
- Tokens for one real question: 294.3K tokens (12th of 13), 1 run
- Team build: hidden tests passed: Not measured

Method: https://adecomparison.com/method/

Rank by must-haves met, then preferences, then name. Read each finding's checked date. Findings stay published after their scheduled recheck. Checks over six months old may be outdated.

## claude-code:ACC-01

Finding: Claude Code has CLI, Desktop, IDE, browser and mobile clients. Execution is local or cloud by surface; mobile is a client only. Desktop Linux is beta; 3P Desktop has separately documented provider and Console-key routes.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Pro, Max, Ultra, Team and Enterprise. Desktop, web and mobile need a claude.ai subscription, except the Desktop exception: gateway routing can be configured in the app or by an administrator; Enterprise deployments can route Desktop to Google Cloud Agent Platform or a gateway via managed settings; and Claude Desktop on 3P runs the Code tab on Amazon Bedrock, Google Cloud Agent Platform, Microsoft Foundry or a self-hosted LLM gateway (feature-availability).
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Capabilities and authentication differ across CLI, Desktop, IDE, mobile and cloud.
No installed-client or authenticated scenario was run.
Desktop 3P supports the documented Console-key route; mobile/cloud still require eligible claude.ai accounts.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/platforms
Locator: Where to run Claude Code; Work when you are away from your terminal
Finding: Compares CLI, Desktop, VS Code, JetBrains, Web and Mobile; CLI is the most complete surface (scripting and Agent SDK are CLI-only); Dispatch, Remote Control, Channels, Slack, self-hosted environments and scheduled tasks listed with where Claude runs and plan requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/setup
Locator: System requirements
Finding: CLI runs on macOS 13+, Windows 10 1809+/Server 2019+, Ubuntu 20.04+, Debian 10+, Alpine 3.19+; 4 GB+ RAM; x64 or ARM64; internet required; Bash, Zsh, PowerShell or CMD.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/desktop
Locator: Sessions; Choose a permission mode; Work in parallel; Dispatch; Coming from the CLI
Finding: Desktop Code tab has parallel sessions with Git isolation, diff review and comments, app preview, integrated terminal, permission modes, scheduled tasks and Dispatch from phone; needs a paid subscription; Desktop does not accept a Console API key directly.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/desktop-linux
Locator: Requirements; What is not in the Linux beta yet
Finding: Linux desktop app is beta on Ubuntu 22.04+/Debian 12+; Fedora/Arch should use the CLI; signs in with claude.ai subscription or SSO, not a Console API key.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/vs-code
Locator: Requirements; Permission modes; Git integration
Finding: VS Code 1.94.0+; extension version equals bundled Claude Code version; inline diffs, plan review, permission mode indicator; also installable in Cursor; shares ~/.claude settings with the CLI.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/jetbrains
Locator: Supported IDEs; Installation; Features
Finding: Plugin runs the separately installed claude CLI in the IDE terminal and adds diff viewing, selection and diagnostic sharing; it does not bundle the CLI.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/sandboxing
Locator: What the sandbox restricts; What runs outside the sandbox
Finding: OS-enforced sandbox for shell commands on macOS, Linux and WSL2 (off by default; native Windows runs unsandboxed); built on open source sandbox-runtime; reads allowed broadly by default, writes limited to working directory, network via domain-allowlist proxy; file tools, MCP servers and hooks run outside it.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/feature-availability
Locator: Availability by model provider; Features that require a Claude subscription
Finding: Local CLI features work across supported providers. Desktop is an exception: gateway/3P deployments can run Code on Bedrock, Google Agent Platform, Foundry, self-hosted gateways and the Claude API with Console key. Subscription-only features have separate plan restrictions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/settings-reference
Locator: Remote, desktop, and notifications
Finding: Settings include agentPushNotifEnabled, inputNeededNotifEnabled, preferredNotifChannel, remoteControlAtStartup, cleanupPeriodDays, autoMemoryEnabled and sandbox keys.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/remote-control
Locator: Introduction; Connect from another device; Requirements
Finding: Remote Control connects claude.ai/code or the mobile apps to a local Claude session. Users can send messages from the terminal, browser and phone; the local host must remain running. Eligible Claude subscriptions are required.
Retrieved: 2026-11-13

## claude-code:ACC-02

Finding: Claude Code embeds its own harness and runs Claude models only. Providers: Anthropic (subscription or Console), Amazon Bedrock, Google Cloud Agent Platform, Microsoft Foundry, Claude Platform on AWS, and gateways that expose an Anthropic-format API. Anthropic does not support routing to non-Claude models and the docs state that workers in every multi-agent mode are Claude sessions.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Scope is the Claude Code harness; unofficial third-party routing to other model families is not covered by the documentation and was not tested.

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/headless
Locator: Basic usage; Bare mode; Structured output
Finding: claude -p runs non-interactively with exit codes, --output-format json/stream-json, --bare, --allowedTools; Agent SDK for Python/TypeScript.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-sdk/overview
Locator: Compare the Agent SDK to other Claude tools; Capabilities
Finding: Python and TypeScript libraries run the Claude Code agent loop with built-in tools, hooks, subagents, MCP, permissions and sessions; non-Python/TS languages can drive the CLI as a subprocess.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/feature-availability
Locator: Availability by model provider; Features that require a Claude subscription
Finding: Local CLI features work across supported providers. Desktop is an exception: gateway/3P deployments can run Code on Bedrock, Google Agent Platform, Foundry, self-hosted gateways and the Claude API with Console key. Subscription-only features have separate plan restrictions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/llm-gateway
Locator: What a gateway provides; Roll out a gateway
Finding: Any gateway exposing a supported Anthropic API format works; Anthropic does not support routing Claude Code to non-Claude models through any gateway.
Retrieved: 2026-11-07

## claude-code:ACC-03

Finding: Pro, Max, Ultra, Team and Enterprise include Claude Code. Console/API and third-party cloud-provider routes are separate. Free excludes Code; platform API credits bundled with Max/Ultra cannot pay Code usage.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Pro, Max, Ultra, Team, Enterprise, Console API, third-party providers
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Local-model statement is the vendor support position; unofficial base-URL workarounds were not tested.
No trial or promotional pricing beyond the public pricing page was checked.
Two Console sign-ins without API keys are not isolated by config directory (authentication page); the account_switching atom is partial for that case.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/channels
Locator: Overview; Enterprise controls
Finding: Research preview. MCP-based channels push Telegram/Discord/iMessage or webhook events into an open session; requires claude.ai login or Console key; not on Bedrock/Agent Platform/Foundry; Team/Enterprise must enable.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/feature-availability
Locator: Availability by model provider; Features that require a Claude subscription
Finding: Local CLI features work across supported providers. Desktop is an exception: gateway/3P deployments can run Code on Bedrock, Google Agent Platform, Foundry, self-hosted gateways and the Claude API with Console key. Subscription-only features have separate plan restrictions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/llm-gateway
Locator: What a gateway provides; Roll out a gateway
Finding: Any gateway exposing a supported Anthropic API format works; Anthropic does not support routing Claude Code to non-Claude models through any gateway.
Retrieved: 2026-11-07

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Current list prices: Pro $20 monthly or $200/year displayed as $17/month; Max $100 monthly; Ultra $200 monthly. Team 2-150 users: Standard $25 monthly/$20 annual per month, Premium $125/$100. Enterprise $20/seat/month annual, 20-seat minimum, plus API usage. All paid plans include Code; taxes excluded. Max/Ultra platform API credits cannot pay Code usage.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

## claude-code:ACC-04

Finding: CLI onboarding is one install command plus browser login, and works in any directory. Worktree isolation needs git; cloud sessions need GitHub (or a local bundle). Self-hosted runners exist for Team/Enterprise. Remote Control is off by default on Team and Enterprise plans until an Owner enables it.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

Onboarding was read from documentation, not timed hands-on.
Self-hosted environments are public beta and exclude ZDR and HIPAA organizations.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/setup
Locator: System requirements
Finding: CLI runs on macOS 13+, Windows 10 1809+/Server 2019+, Ubuntu 20.04+, Debian 10+, Alpine 3.19+; 4 GB+ RAM; x64 or ARM64; internet required; Bash, Zsh, PowerShell or CMD.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/web-quickstart
Locator: Connect GitHub; first session
Finding: Browser onboarding connects GitHub and sets up a Default cloud environment.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/worktrees
Locator: Start Claude in a worktree; Isolate subagents; Non-git version control
Finding: claude --worktree creates .claude/worktrees/<name> on a new branch; requires a git repository (hooks can replace git logic); subagents can use isolation: worktree.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/github-enterprise-server
Locator: Overview
Finding: Self-hosted GitHub Enterprise Server instances supported for Team and Enterprise plans.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/whats-new/index
Locator: Weeks 32 to 37 (2026)
Finding: Weekly digests list cross-session messaging and self-hosted environments (Week 32, public beta), auto mode default (from August 14 for Pro/Max/Team), Desktop resume of CLI sessions (Week 35), Fable 5.1 (Week 36); no future-dated commitments found.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

## claude-code:ACC-05

Finding: Programmable access is strong on the CLI: claude -p with JSON output, Python/TypeScript Agent SDK, claude mcp serve and HTTP/webhook triggers for routines. The routine /fire endpoint is experimental and several surfaces are research preview.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

API/webhook coverage is limited to routines and the SDK; no hands-on integration test.

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/routines
Locator: Triggers; Add an API trigger; GitHub event; Usage and limits
Finding: Cloud routines are now named cloud scheduled tasks; shared web/Desktop/CLI account records, schedule/API/GitHub triggers, research preview fire endpoint, and per-account execution. /schedule creates them and /routines remains an alias.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/headless
Locator: Basic usage; Bare mode; Structured output
Finding: claude -p runs non-interactively with exit codes, --output-format json/stream-json, --bare, --allowedTools; Agent SDK for Python/TypeScript.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-sdk/overview
Locator: Compare the Agent SDK to other Claude tools; Capabilities
Finding: Python and TypeScript libraries run the Claude Code agent loop with built-in tools, hooks, subagents, MCP, permissions and sessions; non-Python/TS languages can drive the CLI as a subprocess.
Retrieved: 2026-11-07

Source: https://github.com/anthropics/claude-agent-sdk-python
Locator: GitHub repository metadata (license field)
Finding: Public repository for the Python Agent SDK; GitHub reports an MIT license (retrieved via the GitHub repository API on 2026-10-08).
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/channels
Locator: Overview; Enterprise controls
Finding: Research preview. MCP-based channels push Telegram/Discord/iMessage or webhook events into an open session; requires claude.ai login or Console key; not on Bedrock/Agent Platform/Foundry; Team/Enterprise must enable.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/cli-reference
Locator: CLI commands; CLI flags
Finding: Documents claude agents, claude remote-control, claude mcp, claude gateway; flags --worktree, --max-turns and --max-budget-usd (print mode only), --fallback-model, --output-format, --permission-mode, --bare, --channels.
Retrieved: 2026-11-07

## claude-code:ACC-06

Finding: Claude Code is an MCP client (http, sse, stdio, websocket) with OAuth and scoped config, supports plugins/marketplaces, skills, hooks, custom subagents, reads AGENTS.md, and can import Codex, Gemini CLI and Cursor configuration.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Integration ownership (which connectors Anthropic maintains) was not established.
Direct AGENTS.md reading requires Claude Code v2.1.277 or later, and in some sessions Claude cannot read AGENTS.md; the docs suggest importing it from CLAUDE.md there.
v2.1.295 negotiates MCP protocol revision 2026-07-28 by default; MCP_PROTOCOL_NEGOTIATION=legacy opts out (changelog, 2026-10-08). The stored answers for this claim do not address protocol versions.

Source: https://code.claude.com/docs/en/agent-sdk/overview
Locator: Compare the Agent SDK to other Claude tools; Capabilities
Finding: Python and TypeScript libraries run the Claude Code agent loop with built-in tools, hooks, subagents, MCP, permissions and sessions; non-Python/TS languages can drive the CLI as a subprocess.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/skills
Locator: Bundled skills
Finding: Skills follow the Agent Skills open standard; custom commands merged into skills; bundled skills include /loop, /batch, /code-review.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/plugins/overview
Locator: Plugin contents; marketplaces
Finding: Plugins package skills, agents, hooks and MCP servers, installed from marketplaces via /plugin.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/admin-setup
Locator: API providers; managed settings; usage monitoring; cloud sessions admin
Finding: Managed settings delivered via claude.ai admin console, MDM or file; SSO, SCIM and seat assignment at the Claude account level; Owners govern cloud environments and the default environment; org effort limits enforced server-side.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cli-reference
Locator: CLI commands; CLI flags
Finding: Documents claude agents, claude remote-control, claude mcp, claude gateway; flags --worktree, --max-turns and --max-budget-usd (print mode only), --fallback-model, --output-format, --permission-mode, --bare, --channels.
Retrieved: 2026-10-09

## claude-code:ACC-07

Finding: The mobile app is a client for cloud or Remote Control sessions, and IDE extensions wrap or bundle the CLI engine. Desktop framework and inspected-source evidence were not established.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2027-01-07
Plan: Not specified
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Desktop UI framework and CLI implementation language were not established from first-party sources; the public repository holds issues, plugins and the changelog under an all-rights-reserved license, and no source inspection was done.

Source: https://code.claude.com/docs/en/platforms
Locator: Where to run Claude Code; Work when you are away from your terminal
Finding: Compares CLI, Desktop, VS Code, JetBrains, Web and Mobile; CLI is the most complete surface (scripting and Agent SDK are CLI-only); Dispatch, Remote Control, Channels, Slack, self-hosted environments and scheduled tasks listed with where Claude runs and plan requirements.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/desktop
Locator: Sessions; Choose a permission mode; Work in parallel; Dispatch; Coming from the CLI
Finding: Desktop Code tab has parallel sessions with Git isolation, diff review and comments, app preview, integrated terminal, permission modes, scheduled tasks and Dispatch from phone; needs a paid subscription; Desktop does not accept a Console API key directly.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/vs-code
Locator: Requirements; Permission modes; Git integration
Finding: VS Code 1.94.0+; extension version equals bundled Claude Code version; inline diffs, plan review, permission mode indicator; also installable in Cursor; shares ~/.claude settings with the CLI.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/jetbrains
Locator: Supported IDEs; Installation; Features
Finding: Plugin runs the separately installed claude CLI in the IDE terminal and adds diff viewing, selection and diagnostic sharing; it does not bundle the CLI.
Retrieved: 2026-10-09

## claude-code:PLAN-01

Finding: No native issue tracker was found in the documentation, changelog or repository checked; work arrives by prompt, Slack/Claude Tag mentions, GitHub @claude mentions and GitHub-event routines, and external trackers such as Linear or Notion are reached through MCP connectors. Authoritative-tracker, ID and sync semantics depend on the connector and were not established.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

A native issue/backlog tracker was not established; session Task tools are a separate documented checklist; the pages checked do not describe tracker identity or sync.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/platforms
Locator: Where to run Claude Code; Work when you are away from your terminal
Finding: Compares CLI, Desktop, VS Code, JetBrains, Web and Mobile; CLI is the most complete surface (scripting and Agent SDK are CLI-only); Dispatch, Remote Control, Channels, Slack, self-hosted environments and scheduled tasks listed with where Claude runs and plan requirements.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/routines
Locator: Triggers; Add an API trigger; GitHub event; Usage and limits
Finding: Research preview. Cloud routines with schedule, API (/fire endpoint under experimental beta header, claude.ai users only) and GitHub event triggers; Pro/Max/Team/Enterprise; run on Anthropic cloud or self-hosted environment; belong to individual account; hourly/webhook caps apply.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/headless
Locator: Basic usage; Bare mode; Structured output
Finding: claude -p runs non-interactively with exit codes, --output-format json/stream-json, --bare, --allowedTools; Agent SDK for Python/TypeScript.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/github-actions
Locator: Setup; mentions
Finding: claude-code-action runs Claude Code in GitHub workflows from @claude mentions or any GitHub event; turns issues into PRs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/slack
Locator: Setup; plans
Finding: The earlier Slack bot is available on Pro, Max and Ultra only in workspaces not connected to an organization Claude Tag. Team/Enterprise use Claude Tag. The retirement notice effective October 5 applies in an organization-connected workspace using the earlier bot.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/tools-reference
Locator: Task tool availability; TaskCreate, TaskGet, TaskList, TaskUpdate
Finding: Product Task tools create, read, update and delete tasks. They are model-dependent or opt-in since v2.1.268, and provided across models in cloud/background sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/interactive-mode
Locator: Task list
Finding: Task checklists are distinct from running shells/subagents, retain state through compaction and resume, and can share a named on-disk task directory across sessions.
Retrieved: 2026-11-07

## claude-code:PLAN-02

Finding: Plan mode proposes a plan before edits and plans can be saved into the repository; /goal tracks a completion condition. Durable plan objects with revisions and acceptance-criteria editors were not found.

Verification: sourced
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Plan files are written to ~/.claude/plans or a project-relative plansDirectory; plan revision history, amendment tracking and links from plans to implementation were not established.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/scheduled-tasks
Locator: Compare scheduling options; Seven-day expiry; Limitations
Finding: /loop and cron tools are session-scoped (1 minute minimum, recurring tasks expire after 7 days, only fire while Claude Code is running and idle, no catch-up for missed fires); cloud routines minimum 1 hour; Desktop tasks need machine on.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/goal
Locator: Compare ways to keep a session running; How evaluation works
Finding: /goal sets a completion condition that a model checks after each turn; goal clears when met, judged impossible, or on an unfixable error.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/permission-modes
Locator: Available modes; Actions no mode auto-approves
Finding: Modes default/Manual, acceptEdits, plan, auto (classifier), dontAsk, bypassPermissions; deny rules apply in every mode; auto is the built-in starting mode for interactive terminal and VS Code sessions from v2.1.283.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/permission-modes
Locator: Review and approve a plan
Finding: Ctrl+G opens the proposed plan in the default text editor before approval; approval exits plan mode.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/settings-reference
Locator: plansDirectory
Finding: plansDirectory sets where plan mode stores plan files, resolved relative to the project root; unset defaults to ~/.claude/plans, and paths outside the project root fall back to ~/.claude/plans.
Retrieved: 2026-11-07

## claude-code:PLAN-03

Finding: Ordinary TaskUpdate changes task dependencies. Task tools are available by default on selected older model families, can be enabled on other models, and are provided on every model in background and cloud sessions. Experimental agent teams add a shared task list where unresolved dependencies block claiming. Hierarchy and epics were not established.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Agent teams are experimental and off by default; no epic/story hierarchy, task-to-epic movement or identity retention was found.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/tools-reference
Locator: Tool list; Task tool availability; WebFetch and WebSearch
Finding: Lists built-in tools including TaskCreate/TaskUpdate/TaskList, TaskStop, WebFetch, WebSearch, Workflow; task tools are provided by default only on some model families.
Retrieved: 2026-11-07

## claude-code:PLAN-04

Finding: Persistent instructions and memory are documented: CLAUDE.md at managed/user/project/local scope, AGENTS.md reading, skills, imports and rules, and editable per-project auto memory stored locally.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Scope verified from documentation only; no hands-on takeover test.
Direct AGENTS.md reading requires Claude Code v2.1.277 or later, and in some sessions Claude cannot read AGENTS.md; the docs suggest importing it from CLAUDE.md there.

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/skills
Locator: Bundled skills
Finding: Skills follow the Agent Skills open standard; custom commands merged into skills; bundled skills include /loop, /batch, /code-review.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/plugins/overview
Locator: Plugin contents; marketplaces
Finding: Plugins package skills, agents, hooks and MCP servers, installed from marketplaces via /plugin.
Retrieved: 2026-10-09

## claude-code:PLAN-05

Finding: Code understanding relies on agentic search (grep, glob, file reads) plus optional LSP code-intelligence plugins and user-supplied MCP indexes; images can be attached; web search and fetch tools exist. A built-in semantic index and citation or conflict-detection features were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No retrieval-quality or citation test was run.

Source: https://code.claude.com/docs/en/platforms
Locator: Where to run Claude Code; Work when you are away from your terminal
Finding: Compares CLI, Desktop, VS Code, JetBrains, Web and Mobile; CLI is the most complete surface (scripting and Agent SDK are CLI-only); Dispatch, Remote Control, Channels, Slack, self-hosted environments and scheduled tasks listed with where Claude runs and plan requirements.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/desktop
Locator: Sessions; Choose a permission mode; Work in parallel; Dispatch; Coming from the CLI
Finding: Desktop Code tab has parallel sessions with Git isolation, diff review and comments, app preview, integrated terminal, permission modes, scheduled tasks and Dispatch from phone; needs a paid subscription; Desktop does not accept a Console API key directly.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cli-reference
Locator: CLI commands; CLI flags
Finding: Documents claude agents, claude remote-control, claude mcp, claude gateway; flags --worktree, --max-turns and --max-budget-usd (print mode only), --fallback-model, --output-format, --permission-mode, --bare, --channels.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/large-codebases
Locator: Code intelligence plugins; Expose your own index over MCP
Finding: Finding symbols can cost many file reads and grep calls; LSP code intelligence plugins add definition/reference lookup; an organization code search or RAG index can be exposed as an MCP tool.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/common-workflows
Locator: Create pull requests; Work with images
Finding: Claude can create commits and pull requests on request; --from-pr finds the session linked to a PR; images can be pasted or referenced by path; Claude reads files fresh on each tool call.
Retrieved: 2026-10-09

## claude-code:PLAN-06

Finding: Claude Code has persistent native task checklists with optional cross-session lists and experimental team assignment/dependencies. A full issue backlog is not established.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-12-07
Plan: Not specified
Surface: Claude Code CLI
Version: Task tools documented 2026-11-07; model-dependent or opt-in since v2.1.268; team assignment experimental

Task tools are absent by default on newer interactive models unless enabled; this is a checklist rather than a full issue tracker. Team assignment requires experimental teams.

Source: https://code.claude.com/docs/en/tools-reference
Locator: Task tool availability; TaskCreate, TaskGet, TaskList, TaskUpdate
Finding: Product Task tools create, read, update and delete tasks. They are model-dependent or opt-in since v2.1.268, and provided across models in cloud/background sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/interactive-mode
Locator: Task list
Finding: Task checklists are distinct from running shells/subagents, retain state through compaction and resume, and can share a named on-disk task directory across sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-teams
Locator: Assign and claim tasks; Architecture; Permissions
Finding: Experimental teams use pending, in-progress and completed tasks, dependencies, explicit assignment and file-locked self-claims. Task directories remain after sessions end.
Retrieved: 2026-11-07

## claude-code:PLAN-07

Finding: TaskCreate, TaskGet, TaskList and TaskUpdate manage session checklists. They are available by default on selected older model families, can be enabled on other models, and are provided on every model in background and cloud sessions. Experimental agent teams add shared claiming and assignment. A native shared issue tracker was not established.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Agent teams are experimental and disabled by default. Local foreground sessions provide task tools by default only on selected model families; other models can opt in. Background and cloud sessions provide them on every model. These are session task lists, not a native issue tracker.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/interactive-mode
Locator: Task list; Wait for a usage limit to reset
Finding: Task list checklist; automatic continue after a claude.ai usage limit resets in interactive subscription sessions; background Bash commands.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/tools-reference
Locator: Tool list; Task tool availability; WebFetch and WebSearch
Finding: Lists built-in tools including TaskCreate/TaskUpdate/TaskList, TaskStop, WebFetch, WebSearch, Workflow; task tools are provided by default only on some model families.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/tools-reference
Locator: Task tool availability; TaskCreate, TaskGet, TaskList, TaskUpdate
Finding: Product Task tools create, read, update and delete tasks. They are model-dependent or opt-in since v2.1.268, and provided across models in cloud/background sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-teams
Locator: Assign and claim tasks; Architecture; Permissions
Finding: Experimental teams use pending, in-progress and completed tasks, dependencies, explicit assignment and file-locked self-claims. Task directories remain after sessions end.
Retrieved: 2026-11-07

## claude-code:PLAN-08

Finding: Living product specifications were not found as a native feature; instructions files and plans in the repository are the nearest documented mechanisms.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

These are general instruction files, not a specification object with approvals or drift detection.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

## claude-code:COORD-01

Finding: Parallel work is supported locally (multiple terminals, worktrees, subagents, agent view background sessions, Remote Control server mode with default capacity 32) and in the cloud; every parallel session draws on the same subscription limits.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Agent view is research preview; no concurrency stress test was run, so observed_concurrency is unknown.

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-10-09

## claude-code:COORD-02

Finding: Agent teams (experimental) document lead assignment and file-locked task claiming; outside teams, assignment is the user dispatching sessions.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Agent teams are experimental and disabled by default; one team per session.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-teams
Locator: Assign and claim tasks; Architecture; Permissions
Finding: Experimental teams use pending, in-progress and completed tasks, dependencies, explicit assignment and file-locked self-claims. Task directories remain after sessions end.
Retrieved: 2026-11-07

## claude-code:COORD-03

Finding: Delegation is inspectable through subagents (own model, effort, tools), agent teams, cross-session messaging and dynamic workflows. All workers are Claude sessions; agent teams are experimental and off by default.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Agent teams are experimental; in-process teammates are not restored on resume.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/cross-session-messaging
Locator: When to use; Availability
Finding: Claude can list and message other sessions on the same or another machine or in the cloud; v2.1.224+ on macOS/Linux, v2.1.234+ on Windows; messages are text only.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/workflows
Locator: When to use a workflow
Finding: A JavaScript script Claude writes orchestrates many subagents in the background; on all paid plans, API, Bedrock, Agent Platform and Foundry; /workflows lists runs.
Retrieved: 2026-11-07

## claude-code:COORD-04

Finding: Cloud scheduled tasks, formerly routines, support schedule/API/GitHub triggers independently of the laptop. Desktop local schedules and session /loop have separate uptime and expiry rules.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Scheduled tasks: Pro, Max, Ultra, Team, Enterprise
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Routines are research preview with experimental API; no scheduled run was executed.

Source: https://code.claude.com/docs/en/routines
Locator: Triggers; Add an API trigger; GitHub event; Usage and limits
Finding: Cloud routines are now named cloud scheduled tasks; shared web/Desktop/CLI account records, schedule/API/GitHub triggers, research preview fire endpoint, and per-account execution. /schedule creates them and /routines remains an alias.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/scheduled-tasks
Locator: Compare scheduling options; Seven-day expiry; Limitations
Finding: /loop and cron tools are session-scoped (1 minute minimum, recurring tasks expire after 7 days, only fire while Claude Code is running and idle, no catch-up for missed fires); cloud routines minimum 1 hour; Desktop tasks need machine on.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/desktop-scheduled-tasks
Locator: Overview; Compare scheduling options
Finding: Local Desktop scheduled tasks run on the user machine and fire only while the app is open and the computer is awake; remote routines run in the cloud.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/goal
Locator: Compare ways to keep a session running; How evaluation works
Finding: /goal sets a completion condition that a model checks after each turn; goal clears when met, judged impossible, or on an unfixable error.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/channels
Locator: Overview; Enterprise controls
Finding: Research preview. MCP-based channels push Telegram/Discord/iMessage or webhook events into an open session; requires claude.ai login or Console key; not on Bedrock/Agent Platform/Foundry; Team/Enterprise must enable.
Retrieved: 2026-11-07

## claude-code:COORD-05

Finding: In the CLI a human can interrupt, stop, queue steering messages and resume sessions; background sessions show Needs input state; push notifications and hooks surface pending decisions. Some commands differ on mobile.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Pause semantics beyond interrupt/stop, escalation policy and error visibility were not tested.

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/data-usage
Locator: Data training policy; Data retention; Telemetry services; Default behaviors by API provider
Finding: Consumer plans: training on opt-in setting, retention 5 years if allowed else 30 days; commercial plans no training by default, 30-day standard retention, ZDR for qualified Enterprise accounts; local transcripts kept 30 days by default; telemetry/error reporting defaults per provider with env-var opt-outs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/settings-reference
Locator: Remote, desktop, and notifications
Finding: Settings include agentPushNotifEnabled, inputNeededNotifEnabled, preferredNotifChannel, remoteControlAtStartup, cleanupPeriodDays, autoMemoryEnabled and sandbox keys.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/tools-reference
Locator: Tool list; Task tool availability; WebFetch and WebSearch
Finding: Lists built-in tools including TaskCreate/TaskUpdate/TaskList, TaskStop, WebFetch, WebSearch, Workflow; task tools are provided by default only on some model families.
Retrieved: 2026-10-09

## claude-code:COORD-06

Finding: Agent view (research preview) is a terminal fleet screen that groups background sessions under Needs input, Ready for review, Working and Completed, with filters, peek and attach; Desktop has a session sidebar. Dependency graphs and parent-child epic views were not found.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Agent view is research preview and shows only local background sessions; no fixture with many agents was run.

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-11-07

## claude-code:COORD-07

Finding: Quick tasks are one prompt away; larger efforts use subagents, agent teams, projects (cloud beta) or dynamic workflows, configured in natural language or markdown. Agent teams are experimental.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Orchestration surfaces are spread across several preview features; no end-to-end scenario run.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/headless
Locator: Basic usage; Bare mode; Structured output
Finding: claude -p runs non-interactively with exit codes, --output-format json/stream-json, --bare, --allowedTools; Agent SDK for Python/TypeScript.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/workflows
Locator: When to use a workflow
Finding: A JavaScript script Claude writes orchestrates many subagents in the background; on all paid plans, API, Bedrock, Agent Platform and Foundry; /workflows lists runs.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-projects
Locator: Overview; Availability note
Finding: Projects public beta gradually rolls out to Pro/Max/Ultra; not Team/Enterprise. Parallel cloud threads, optionally local through Remote Control. Projects are single-user and not shareable.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/skills
Locator: Bundled skills
Finding: Skills follow the Agent Skills open standard; custom commands merged into skills; bundled skills include /loop, /batch, /code-review.
Retrieved: 2026-11-07

## claude-code:RUN-01

Finding: Claude Code runs locally by default; cloud sessions run on Anthropic VMs (4 vCPU, 16 GB, 30 GB) with no separate compute charge; cloud sessions are available on Pro, Max, Ultra and Team plans, and for Enterprise users with premium seats or Chat + Claude Code seats, optionally on customer-run self-hosted runners (Team/Enterprise beta); Remote Control lets other devices drive a local session.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Cloud: Pro, Max, Ultra, Team, Enterprise; self-hosted: Team, Enterprise (public beta)
Surface: Claude Code CLI; Claude Code web/mobile
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Region and exact idle-expiry durations were not found.

Source: https://code.claude.com/docs/en/overview
Locator: Get started tabs; Terminal install; Desktop app; Web
Finding: Documents terminal (macOS/Linux/WSL/Windows), VS Code, desktop (macOS, Windows x64/ARM64, Ubuntu/Debian beta), web and JetBrains surfaces; most surfaces need a Claude subscription or Console account; CLI, VS Code and JetBrains also support third-party providers.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/platforms
Locator: Where to run Claude Code; Work when you are away from your terminal
Finding: Compares CLI, Desktop, VS Code, JetBrains, Web and Mobile; CLI is the most complete surface (scripting and Agent SDK are CLI-only); Dispatch, Remote Control, Channels, Slack, self-hosted environments and scheduled tasks listed with where Claude runs and plan requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/claude-projects
Locator: Overview; Availability note
Finding: Projects public beta gradually rolls out to Pro/Max/Ultra; not Team/Enterprise. Parallel cloud threads, optionally local through Remote Control. Projects are single-user and not shareable.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/settings-reference
Locator: Remote, desktop, and notifications
Finding: Settings include agentPushNotifEnabled, inputNeededNotifEnabled, preferredNotifChannel, remoteControlAtStartup, cleanupPeriodDays, autoMemoryEnabled and sandbox keys.
Retrieved: 2026-11-07

## claude-code:RUN-02

Finding: Isolation options: git worktrees per session, an OS-level sandbox for shell commands (macOS/Linux/WSL2, off by default), dev containers or the sandbox runtime for the whole process, and a fresh VM per cloud session. The CLI by default runs with the user permissions and file tools outside the sandbox.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

No escape or boundary test was run.

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/worktrees
Locator: Start Claude in a worktree; Isolate subagents; Non-git version control
Finding: claude --worktree creates .claude/worktrees/<name> on a new branch; requires a git repository (hooks can replace git logic); subagents can use isolation: worktree.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sandboxing
Locator: What the sandbox restricts; What runs outside the sandbox
Finding: OS-enforced sandbox for shell commands on macOS, Linux and WSL2 (off by default; native Windows runs unsandboxed); built on open source sandbox-runtime; reads allowed broadly by default, writes limited to working directory, network via domain-allowlist proxy; file tools, MCP servers and hooks run outside it.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sandbox-environments
Locator: Sandbox runtime; dev containers; VMs
Finding: Compares built-in sandbox, sandbox runtime, dev containers, containers and VMs for isolating Claude Code.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/worktrees
Locator: Start Claude in a worktree; Isolate subagents with worktrees; What worktrees share with the main checkout
Finding: The --worktree or -w flag starts a session in a separate Git working directory, by default .claude/worktrees/<name>, on a new worktree-<name> branch. Subagents can use isolation: worktree. Worktrees share repository history and Git metadata; this documents file separation, not a security sandbox. Source check only; no session was run.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/sandbox-environments
Locator: Sandbox runtime; dev containers; VMs
Finding: Compares built-in sandbox, sandbox runtime, dev containers, containers and VMs for isolating Claude Code.
Retrieved: 2026-11-07

## claude-code:RUN-03

Finding: Sessions persist as local transcripts and resume; cloud sessions restore history after VM reclamation but lose background processes; Remote Control reconnects after sleep, and server mode gives up after roughly 10 minutes without network; background local sessions survive sleep but not shutdown.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No crash or restart experiments were run; claims are from documentation.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/scheduled-tasks
Locator: Compare scheduling options; Seven-day expiry; Limitations
Finding: /loop and cron tools are session-scoped (1 minute minimum, recurring tasks expire after 7 days, only fire while Claude Code is running and idle, no catch-up for missed fires); cloud routines minimum 1 hour; Desktop tasks need machine on.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/checkpointing
Locator: How checkpoints work
Finding: Automatic snapshots before each prompt (100 most recent kept), /rewind restores code and/or conversation; file snapshots swept by retention (~30 days by default).
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/data-usage
Locator: Data training policy; Data retention; Telemetry services; Default behaviors by API provider
Finding: Consumer plans: training on opt-in setting, retention 5 years if allowed else 30 days; commercial plans no training by default, 30-day standard retention, ZDR for qualified Enterprise accounts; local transcripts kept 30 days by default; telemetry/error reporting defaults per provider with env-var opt-outs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/changelog
Locator: Version 2.1.294 (October 8, 2026) and recent entries
Finding: Newest entry is v2.1.295 (2026-10-08), then v2.1.294 (2026-10-08); v2.1.287 to v2.1.295 are dated Oct 1 to 8. v2.1.295 negotiates MCP protocol 2026-07-28 by default; MCP_PROTOCOL_NEGOTIATION=legacy opts out.
Retrieved: 2026-10-09

## claude-code:RUN-04

Finding: Cloud environments offer setup scripts, a filesystem-snapshot cache, network access levels and pre-installed toolchains; local runs inherit the host. Process state, ports and automatic repair are not covered.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code web/mobile
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

Cloud-only findings; the local CLI inherits whatever the host provides.

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/worktrees
Locator: Start Claude in a worktree; Isolate subagents; Non-git version control
Finding: claude --worktree creates .claude/worktrees/<name> on a new branch; requires a git repository (hooks can replace git logic); subagents can use isolation: worktree.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

## claude-code:RUN-05

Finding: Limits include print-mode --max-turns and --max-budget-usd (client-side estimate), subagent maxTurns, tool timeouts, plan usage limits with automatic continue, organization spend limits, 10 automatic retries and --fallback-model.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

--max-budget-usd and --max-turns apply to print mode only; subagent maxTurns caps individual subagents in any session.

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/sub-agents
Locator: Built-in subagents; Supported frontmatter fields; Choose a model; Run subagents in foreground or background
Finding: Subagents have their own context, tools, permissions, model, effort, maxTurns and optional worktree isolation; background subagents run concurrently; results return to the spawning conversation.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/cli-reference
Locator: CLI commands; CLI flags
Finding: Documents claude agents, claude remote-control, claude mcp, claude gateway; flags --worktree, --max-turns and --max-budget-usd (print mode only), --fallback-model, --output-format, --permission-mode, --bare, --channels.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/interactive-mode
Locator: Task list; Wait for a usage limit to reset
Finding: Task list checklist; automatic continue after a claude.ai usage limit resets in interactive subscription sessions; background Bash commands.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/errors
Locator: Automatic retries
Finding: Claude Code retries transient failures up to 10 times with exponential backoff; partial-response failures are not always retried.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

## claude-code:RUN-06

Finding: Operational statistics are available via /usage, /insights, OpenTelemetry metrics, logs and traces (beta), and the Team/Enterprise analytics dashboard; failure reasons are documented in the error reference.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No accuracy or latency test was run.

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/monitoring-usage
Locator: Quick start; Metrics; Events; Traces
Finding: OpenTelemetry export of metrics (sessions, cost, tokens, lines of code, commits, PRs, active time), events and optional traces.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/analytics
Locator: Access analytics for Team and Enterprise
Finding: Team/Enterprise dashboard at claude.ai/analytics/claude-code with usage and contribution metrics, leaderboard and CSV export; Console dashboard for API customers.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/errors
Locator: Automatic retries
Finding: Claude Code retries transient failures up to 10 times with exponential backoff; partial-response failures are not always retried.
Retrieved: 2026-10-09

## claude-code:RUN-07

Finding: Cost views: /usage session cost (client-side estimate at list price unless an administrator's managed modelPricing table sets contracted rates), plan usage bars, OTel cost and token metrics, Console and Team/Enterprise dashboards; no separate compute charge for cloud VMs. Estimates can differ from the bill.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Pro, Max, Team, Enterprise, Console API
Surface: Claude Code CLI; Claude Code web/mobile
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Costs reconcile against provider billing only outside Claude Code.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/monitoring-usage
Locator: Quick start; Metrics; Events; Traces
Finding: OpenTelemetry export of metrics (sessions, cost, tokens, lines of code, commits, PRs, active time), events and optional traces.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/analytics
Locator: Access analytics for Team and Enterprise
Finding: Team/Enterprise dashboard at claude.ai/analytics/claude-code with usage and contribution metrics, leaderboard and CSV export; Console dashboard for API customers.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

## claude-code:RUN-08

Finding: Compute-aware scheduling is limited: Remote Control and self-hosted runners cap concurrency, Bash commands have timeouts and can be cancelled, and subscription limits are shared. Workflow fan-outs run up to 16 agents by default and fewer on CPU-limited hosts, configurable from 1 to 256 (v2.1.269 or later). RAM-aware queues, check deduplication and fairness were not found.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

Workflow concurrency adapts to available CPUs; RAM-aware scheduling was not established.

Source: https://code.claude.com/docs/en/remote-control
Locator: Requirements; Start a Remote Control session; Limitations; Connection and security
Finding: Drive a local CLI/Desktop/VS Code session from claude.ai/code or the mobile app; machine must stay on; Pro/Max/Team/Enterprise only (no API keys, Bedrock/Vertex/Foundry, gateways); server mode default capacity 32; transcript stored on Anthropic servers while connected.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/workflows
Locator: When to use a workflow
Finding: A JavaScript script Claude writes orchestrates many subagents in the background; on all paid plans, API, Bedrock, Agent Platform and Foundry; /workflows lists runs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/interactive-mode
Locator: Task list; Wait for a usage limit to reset
Finding: Task list checklist; automatic continue after a claude.ai usage limit resets in interactive subscription sessions; background Bash commands.
Retrieved: 2026-10-09

Source: https://support.claude.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro/Max usage limits are shared across Claude and Claude Code; options at limit include usage credits, Console API credits or waiting; monthly API credits included with Max cannot be used in Claude Code; ANTHROPIC_API_KEY overrides subscription auth and causes API charges.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/cloud-environments
Locator: Installed tools; Time limits; Setup scripts; Environment caching; Network access
Finding: Anthropic-hosted sessions use fresh Ubuntu 24.04 x86_64 VMs with approximate ceilings of 4 vCPUs, 16 GB RAM and 30 GB disk. Setup under roughly five minutes is cached as a filesystem snapshot, expiring after roughly seven days. Foreground Bash defaults to two minutes, up to ten; timed-out commands move to background except sleep-prefixed commands. Unattended background commands have a separate time limit. Idle VMs pause and can be reclaimed. Network secrets are available on Pro, Max, Ultra, Team and Enterprise, with role/configuration requirements.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/workflows
Locator: How a workflow runs > Behavior and limits
Finding: Workflows run up to 16 concurrent agents by default and fewer on hosts or containers with fewer CPUs; CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS accepts 1 to 256 and needs v2.1.269 or later.
Retrieved: 2026-11-07

## claude-code:TEAM-01

Finding: Live multi-human collaboration is limited. Cloud sessions can be shared by link (Team visibility or public for Pro/Max) but recipients see the latest state without real-time updates; multiple terminals on one session interleave a single transcript.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Team/Enterprise: Private or Team visibility; Pro/Max: Private or Public
Surface: Claude Code CLI; Claude Code web/mobile
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No presence, co-prompting or comment threads were found for the CLI; no multi-user test was run.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/artifacts
Locator: Overview; Availability
Finding: Artifacts publish live interactive pages from a session to a private claude.ai URL that can be shared with the organization or a public link; Pro, Max, Team and Enterprise with /login.
Retrieved: 2026-10-09

## claude-code:TEAM-02

Finding: Roles and access are managed at the Claude account level: Owners and admins configure cloud environments and settings; Team adds SSO and central billing; Enterprise adds role-based access, SCIM, audit logs and spend limits.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Team, Enterprise
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Project-level and task-level roles were not found; role definitions were not read in detail.

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/admin-setup
Locator: API providers; managed settings; usage monitoring; cloud sessions admin
Finding: Managed settings delivered via claude.ai admin console, MDM or file; SSO, SCIM and seat assignment at the Claude account level; Owners govern cloud environments and the default environment; org effort limits enforced server-side.
Retrieved: 2026-10-09

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Free $0 (Claude Code: No); Pro $17/month annual or $20 monthly (Claude Code: Yes); Max from $100/month (5x or 20x Pro usage); Team standard seat $20 annual/$25 monthly, premium seat $100 annual/$125 monthly (Claude Code included); Enterprise US$20/seat/month plus usage at API rates; usage limits apply; prices exclude tax; model training opt-out on consumer plans, none by default on Team/Enterprise.
Retrieved: 2026-10-09

## claude-code:TEAM-03

Finding: Approvals are per-user permission prompts governed by permission modes plus allow/ask/deny rules, hooks and organization managed settings. There is no separate approver identity; mobile and Remote Control sessions can answer prompts.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Permission behaviour was read from documentation and not exercised.

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sandboxing
Locator: What the sandbox restricts; What runs outside the sandbox
Finding: OS-enforced sandbox for shell commands on macOS, Linux and WSL2 (off by default; native Windows runs unsandboxed); built on open source sandbox-runtime; reads allowed broadly by default, writes limited to working directory, network via domain-allowlist proxy; file tools, MCP servers and hooks run outside it.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/permission-modes
Locator: Available modes; Actions no mode auto-approves
Finding: Modes default/Manual, acceptEdits, plan, auto (classifier), dontAsk, bypassPermissions; deny rules apply in every mode; auto is the built-in starting mode for interactive terminal and VS Code sessions from v2.1.283.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/mcp
Locator: Installing MCP servers; Authenticate with remote MCP servers; Use Claude Code as an MCP server; Scopes
Finding: MCP client supports http, sse, stdio and websocket servers with OAuth 2.0, local/project/user scopes and claude.ai connectors; claude mcp serve exposes Claude Code as a stdio MCP server; channels push events.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/admin-setup
Locator: API providers; managed settings; usage monitoring; cloud sessions admin
Finding: Managed settings delivered via claude.ai admin console, MDM or file; SSO, SCIM and seat assignment at the Claude account level; Owners govern cloud environments and the default environment; org effort limits enforced server-side.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/permissions
Locator: Manage permissions; permission rule syntax; protected paths
Finding: Allow, ask and deny rules for tools and Bash command patterns (for example git commit and git push patterns) layered on permission modes; deny rules apply in every mode; managed settings take precedence over local settings.
Retrieved: 2026-10-09

## claude-code:TEAM-04

Finding: History comes from local session transcripts, checkpoints, OpenTelemetry events and, on Enterprise, audit logs. CLI transcripts default to 30 days of local retention, and Desktop and Cowork sessions are exempt by default (data-usage page); cross-user persistence and integrity guarantees were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Enterprise audit logs
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Audit log contents for Enterprise were not read.

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/checkpointing
Locator: How checkpoints work
Finding: Automatic snapshots before each prompt (100 most recent kept), /rewind restores code and/or conversation; file snapshots swept by retention (~30 days by default).
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/monitoring-usage
Locator: Quick start; Metrics; Events; Traces
Finding: OpenTelemetry export of metrics (sessions, cost, tokens, lines of code, commits, PRs, active time), events and optional traces.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/analytics
Locator: Access analytics for Team and Enterprise
Finding: Team/Enterprise dashboard at claude.ai/analytics/claude-code with usage and contribution metrics, leaderboard and CSV export; Console dashboard for API customers.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/data-usage
Locator: Data training policy; Data retention; Telemetry services; Default behaviors by API provider
Finding: Consumer plans: training on opt-in setting, retention 5 years if allowed else 30 days; commercial plans no training by default, 30-day standard retention, ZDR for qualified Enterprise accounts; local transcripts kept 30 days by default; telemetry/error reporting defaults per provider with env-var opt-outs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

## claude-code:TEAM-05

Finding: Review is available through /diff, Desktop diff view with inline comments, local /code-review, and managed GitHub PR Code Review (Team/Enterprise) whose findings do not block merges. The Team and Enterprise review feature is not available for organizations with Zero Data Retention or the HIPAA configuration (code-review page). Batch review UX was not established.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code desktop
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

No review experience test; effort, keyboard and narrow-screen behaviour were not examined.
Managed Code Review also excludes organizations using customer-managed encryption keys; the current page no longer labels it research preview.

Source: https://code.claude.com/docs/en/desktop
Locator: Sessions; Choose a permission mode; Work in parallel; Dispatch; Coming from the CLI
Finding: Desktop Code tab has parallel sessions with Git isolation, diff review and comments, app preview, integrated terminal, permission modes, scheduled tasks and Dispatch from phone; needs a paid subscription; Desktop does not accept a Console API key directly.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/code-review
Locator: Overview; local review
Finding: Managed GitHub PR Code Review is available to eligible Team/Enterprise organizations and excludes ZDR/HIPAA and customer-managed encryption keys. Inline findings and a non-failing check do not approve or block merges. The current page does not label the service research preview. Local /code-review is a separate route.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-11-07

## claude-code:TEAM-06

Finding: Takeover is mostly same-user: teleport and Remote Control move a session between one person own devices; colleagues can open shared cloud session links to see the latest state. Transfer of ownership between users was not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No colleague takeover scenario was run.

Source: https://code.claude.com/docs/en/mobile
Locator: Work from your phone; Get push notifications; Limitations
Finding: The Claude iOS/Android app is a client for cloud sessions, projects, Remote Control and Dispatch (no code runs on the phone). Push notifications when Remote Control is active; local-only commands such as /plugin and /resume do not work; Bypass permissions cannot be selected from the app.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

## claude-code:SHIP-01

Finding: Outputs include diffs (/diff, Desktop diff view), live Artifact pages published to claude.ai, Desktop app preview, and session links to PRs. An artifact gallery lists created artifacts. Retention of test outputs and recordings was not established.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Artifacts: Pro, Max, Team, Enterprise
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Artifact availability requires a claude.ai login and plan; no output retention test was run.

Source: https://code.claude.com/docs/en/desktop
Locator: Sessions; Choose a permission mode; Work in parallel; Dispatch; Coming from the CLI
Finding: Desktop Code tab has parallel sessions with Git isolation, diff review and comments, app preview, integrated terminal, permission modes, scheduled tasks and Dispatch from phone; needs a paid subscription; Desktop does not accept a Console API key directly.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/jetbrains
Locator: Supported IDEs; Installation; Features
Finding: Plugin runs the separately installed claude CLI in the IDE terminal and adds diff viewing, selection and diagnostic sharing; it does not bundle the CLI.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/common-workflows
Locator: Create pull requests; Work with images
Finding: Claude can create commits and pull requests on request; --from-pr finds the session linked to a PR; images can be pasted or referenced by path; Claude reads files fresh on each tool call.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/artifacts
Locator: Overview; Availability
Finding: Artifacts publish live interactive pages from a session to a private claude.ai URL that can be shared with the organization or a public link; Pro, Max, Team and Enterprise with /login.
Retrieved: 2026-10-09

## claude-code:SHIP-02

Finding: Verification is policy you configure: hooks run commands and can block actions or task completion (exit code 2), /verify builds and drives the app, /goal re-checks a completion condition, and /code-review and managed Code Review provide independent review. Managed settings can restrict which hooks run (allowManagedHooksOnly), but the docs do not state that a check must pass before delivery.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

No gate was exercised.

Source: https://code.claude.com/docs/en/agent-teams
Locator: Enablement; shared task list; Assign and claim tasks; Limitations
Finding: Experimental, disabled by default (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). Lead plus independent teammates with own context, direct messaging, shared task list with pending/in-progress/completed states and task dependencies, file-locked claiming; limitations include no in-process teammate resume, one team per session, status lag.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/goal
Locator: Compare ways to keep a session running; How evaluation works
Finding: /goal sets a completion condition that a model checks after each turn; goal clears when met, judged impossible, or on an unfixable error.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/hooks-guide
Locator: What you can automate; Hook types; Hooks and permission modes
Finding: Hooks fire on events such as PreToolUse, PostToolUse, PermissionRequest, Notification, Stop, SubagentStop, SessionStart/End, TaskCreated/Completed, WorktreeCreate and can be commands, prompts, agents or HTTP calls; can block or auto-approve actions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/admin-setup
Locator: API providers; managed settings; usage monitoring; cloud sessions admin
Finding: Managed settings delivered via claude.ai admin console, MDM or file; SSO, SCIM and seat assignment at the Claude account level; Owners govern cloud environments and the default environment; org effort limits enforced server-side.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/github-actions
Locator: Setup; mentions
Finding: claude-code-action runs Claude Code in GitHub workflows from @claude mentions or any GitHub event; turns issues into PRs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/permissions
Locator: Manage permissions; permission rule syntax; protected paths
Finding: Allow, ask and deny rules for tools and Bash command patterns (for example git commit and git push patterns) layered on permission modes; deny rules apply in every mode; managed settings take precedence over local settings.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/code-review
Locator: Overview; local review
Finding: Managed GitHub PR Code Review is available to eligible Team/Enterprise organizations and excludes ZDR/HIPAA and customer-managed encryption keys. Inline findings and a non-failing check do not approve or block merges. The current page does not label the service research preview. Local /code-review is a separate route.
Retrieved: 2026-11-07

## claude-code:SHIP-03

Finding: The CLI supports git workflows: it can create commits and pull requests on request, run in isolated worktrees, link sessions to PRs, and (cloud) auto-fix CI failures and review comments on GitHub.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI; Claude Code web/mobile
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

GitHub is the supported PR host for cloud features.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/worktrees
Locator: Start Claude in a worktree; Isolate subagents; Non-git version control
Finding: claude --worktree creates .claude/worktrees/<name> on a new branch; requires a git repository (hooks can replace git logic); subagents can use isolation: worktree.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/code-review
Locator: Overview; local review
Finding: Research preview managed GitHub PR review for Team/Enterprise (not with ZDR/HIPAA); findings are inline comments and do not approve or block PRs; /code-review reviews a local diff on other plans.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/common-workflows
Locator: Create pull requests; Work with images
Finding: Claude can create commits and pull requests on request; --from-pr finds the session linked to a PR; images can be pasted or referenced by path; Claude reads files fresh on each tool call.
Retrieved: 2026-10-09

## claude-code:SHIP-04

Finding: Merge and deployment authority is delegated to git/CLI permissions: Bash allow, ask or deny rules can gate push and merge commands. Dedicated approval, release and deploy workflows were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No merge or deploy flow was tested.

Source: https://code.claude.com/docs/en/permissions
Locator: Manage permissions; permission rule syntax; protected paths
Finding: Allow, ask and deny rules for tools and Bash command patterns (for example git commit and git push patterns) layered on permission modes; deny rules apply in every mode; managed settings take precedence over local settings.
Retrieved: 2026-10-09

## claude-code:SHIP-05

Finding: Completion evidence is session links to PRs and OTel commit/PR counts; rollback is via /rewind checkpoints on local files. Build and deployment links and stale-approval invalidation were not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Local rewind is not a release rollback.

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/checkpointing
Locator: How checkpoints work
Finding: Automatic snapshots before each prompt (100 most recent kept), /rewind restores code and/or conversation; file snapshots swept by retention (~30 days by default).
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/monitoring-usage
Locator: Quick start; Metrics; Events; Traces
Finding: OpenTelemetry export of metrics (sessions, cost, tokens, lines of code, commits, PRs, active time), events and optional traces.
Retrieved: 2026-10-09

## claude-code:SHIP-06

Finding: High-volume shipping features are limited: agent view summaries, /batch splitting a change into worktree subagents, and Code Review severity tags. A review queue with exact-revision approval was not found.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Not a review queue.

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/agent-view
Locator: Quick start; How file edits are isolated; Limitations; Version history
Finding: Research preview. claude agents dispatches and monitors background local sessions; sessions move into git worktrees before editing; background sessions use subscription quota like interactive ones, stop if the machine shuts down; claude agents --json output.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/code-review
Locator: Overview; local review
Finding: Research preview managed GitHub PR review for Team/Enterprise (not with ZDR/HIPAA); findings are inline comments and do not approve or block PRs; /code-review reviews a local diff on other plans.
Retrieved: 2026-10-09

## claude-code:OWN-01

Finding: Claude Code is proprietary: the public repository LICENSE.md states "© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service." The sandbox runtime (Apache-2.0) and the Python Agent SDK (MIT, per GitHub) are open source components.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2027-01-07
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

No source-inspected evidence recorded.
Head commit 71cdddec623889d38af14b7a489670a03186f659 was not re-verified on 2026-10-09; the repository main page did not expose it.

Source: https://github.com/anthropics/claude-agent-sdk-python
Locator: GitHub repository metadata (license field)
Finding: Public repository for the Python Agent SDK; GitHub reports an MIT license (retrieved via the GitHub repository API on 2026-10-08).
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/changelog
Locator: Version 2.1.294 (October 8, 2026) and recent entries
Finding: Newest entry is v2.1.295 (2026-10-08), then v2.1.294 (2026-10-08); v2.1.287 to v2.1.295 are dated Oct 1 to 8. v2.1.295 negotiates MCP protocol 2026-07-28 by default; MCP_PROTOCOL_NEGOTIATION=legacy opts out.
Retrieved: 2026-10-09

Source: https://github.com/anthropics/claude-code
Locator: Repository metadata, LICENSE.md, releases and issues
Finding: Releases page marks v2.1.295 (08 Oct 19:48) as Latest, above stored v2.1.294. Issues tab shows 13,889 open (stored 14,591). Main page shows 149.8k stars; no license name or head commit visible.
Retrieved: 2026-10-09

Source: https://github.com/anthropics/sandbox-runtime
Locator: GitHub repository metadata (license field)
Finding: Open source sandbox-runtime package used for the built-in sandbox; GitHub reports Apache-2.0.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/legal-and-compliance
Locator: Legal agreements; Can customers offer Claude Code in their products?
Finding: Use is governed by Commercial Terms (Team, Enterprise, API) or Consumer Terms (Free, Pro, Max); products embedding Claude Code must not modify the binary or resell usage on end users behalf.
Retrieved: 2026-10-09

Source: https://raw.githubusercontent.com/anthropics/claude-code/main/LICENSE.md
Locator: Full file text (single line)
Finding: The whole file reads: "© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service." No open-source license grant appears in the file.
Retrieved: 2026-10-09

Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.295
Locator: release header; tag target
Finding: Release v2.1.295 published 08 Oct 19:48 and marked Latest; tag resolves to commit 602df92.
Retrieved: 2026-10-09

## claude-code:OWN-02

Finding: The installed Claude Code CLI can run its full agent loop, tools and context management on user-controlled infrastructure, including scripts and CI/CD. Model inference/authentication still uses Anthropic or a supported provider. This does not self-host the Claude web/desktop cloud control plane; cloud self-hosted runners are a separate worker feature.

Verification: sourced
Checked: 2026-11-13
Scheduled recheck: 2027-05-12
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code CLI headless documentation checked 2026-11-13 for local-core self-hosting. Other atomic answers retain their individually dated source scope.

The mapping covers the local CLI core only; hosted cloud/desktop service self-hosting is not established.

Source: https://code.claude.com/docs/en/setup
Locator: System requirements
Finding: CLI runs on macOS 13+, Windows 10 1809+/Server 2019+, Ubuntu 20.04+, Debian 10+, Alpine 3.19+; 4 GB+ RAM; x64 or ARM64; internet required; Bash, Zsh, PowerShell or CMD.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/memory
Locator: CLAUDE.md files; AGENTS.md; Auto memory; View and edit with /memory
Finding: CLAUDE.md at managed/user/project/local scope with imports and .claude/rules; AGENTS.md read when no CLAUDE.md (v2.1.277+); auto memory stored per project under ~/.claude/projects/<project>/memory/ as editable markdown; /memory browses and toggles.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/sessions
Locator: Resume a session; Export and locate session data
Finding: Sessions saved continuously as local JSONL transcripts; --continue, --resume, --from-pr, /branch; running the same session in two terminals interleaves one transcript; /export writes plain text.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/feature-availability
Locator: Availability by model provider; Features that require a Claude subscription
Finding: CLI, SDK, IDE extensions, subagents, hooks, skills, memory, plugins, MCP, checkpoints, sandboxing, workflows work on every provider; cloud sessions, mobile, Slack, Desktop need a claude.ai subscription.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/commands
Locator: All commands
Finding: Lists /usage, /export, /import (Codex, Gemini CLI, Cursor), /rewind, /branch, /fork, /goal, /loop, /schedule, /teleport, /desktop, /code-review, /security-review, /install-github-app, /web-setup, /tasks, /sandbox, /diff.
Retrieved: 2026-10-09

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Free $0 (Claude Code: No); Pro $17/month annual or $20 monthly (Claude Code: Yes); Max from $100/month (5x or 20x Pro usage); Team standard seat $20 annual/$25 monthly, premium seat $100 annual/$125 monthly (Claude Code included); Enterprise US$20/seat/month plus usage at API rates; usage limits apply; prices exclude tax; model training opt-out on consumer plans, none by default on Team/Enterprise.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/headless
Locator: Agent SDK introduction; CLI scripts and CI/CD; Basic usage
Finding: The CLI exposes the same agent loop, tools and context management as Claude Code through claude -p for scripts and CI/CD. The installed local core runs with model-provider credentials; this is more than a remote execution worker for the hosted web product.
Retrieved: 2026-11-13

## claude-code:OWN-03

Finding: Prompts and outputs go to Anthropic (or the chosen provider) over TLS 1.2+; consumer plans train on data only when the setting is on, commercial plans do not by default; standard retention is 30 days for commercial and 30 days or 5 years for consumers. Usage metrics default on for the Claude API and off on Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry; error reports default on only for Pro, Max and Ultra sign-ins on v2.1.198+; both have documented opt-outs.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Retention and training differ by plan
Surface: Claude Code CLI
Version: Current first-party documentation read 2026-11-07; changelog through v2.1.319

Plan-specific; Remote Control and cloud flows differ.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-10-09

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Free $0 (Claude Code: No); Pro $17/month annual or $20 monthly (Claude Code: Yes); Max from $100/month (5x or 20x Pro usage); Team standard seat $20 annual/$25 monthly, premium seat $100 annual/$125 monthly (Claude Code included); Enterprise US$20/seat/month plus usage at API rates; usage limits apply; prices exclude tax; model training opt-out on consumer plans, none by default on Team/Enterprise.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/data-usage
Locator: Data training policy; Data retention; Telemetry services; Default behaviors by API provider
Finding: Current data policy states opt-in consumer training and five-year versus thirty-day retention by choice; commercial no training by default and thirty-day standard retention. Local plaintext CLI transcripts default to thirty days, with Desktop/Cowork exemptions. Metrics and error reports have separate opt-outs. Error reports default on for direct Claude API Pro, Max and Ultra sign-ins on v2.1.198+, otherwise off. Provider-managed-host metrics are an explicit exception to third-party-provider default-off behavior.
Retrieved: 2026-11-07

## claude-code:OWN-04

Finding: Pro $20/month or $200/year; Max $100/month; Ultra $200/month. Team seats $25/$125 monthly or $20/$100 monthly-equivalent annual. Enterprise $20/seat/month annual with 20-seat minimum, plus API usage.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Pro, Max, Ultra, Team, Enterprise
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

No workload, billing or authenticated client test was performed; unestablished details remain unknown.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Current list prices: Pro $20 monthly or $200/year displayed as $17/month; Max $100 monthly; Ultra $200 monthly. Team 2-150 users: Standard $25 monthly/$20 annual per month, Premium $125/$100. Enterprise $20/seat/month annual, 20-seat minimum, plus API usage. All paid plans include Code; taxes excluded. Max/Ultra platform API credits cannot pay Code usage.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

## claude-code:OWN-05

Finding: Fixed subscription examples: one Pro $20/month or $200/year; two Team Standard seats $50/month or $480/year; five $125/month or $1,200/year. Variable workload costs remain unknown.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

No workload, billing or authenticated client test was performed; unestablished details remain unknown.

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Current list prices: Pro $20 monthly or $200/year displayed as $17/month; Max $100 monthly; Ultra $200 monthly. Team 2-150 users: Standard $25 monthly/$20 annual per month, Premium $125/$100. Enterprise $20/seat/month annual, 20-seat minimum, plus API usage. All paid plans include Code; taxes excluded. Max/Ultra platform API credits cannot pay Code usage.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

## claude-code:OWN-06

Finding: Current changelog lists v2.1.319. Feature preview/beta flags, country/account access and status history are documented separately; no response-time guarantee was established.

Verification: sourced
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

No SLA or support-response data was found.

Source: https://code.claude.com/docs/en/setup
Locator: System requirements
Finding: CLI runs on macOS 13+, Windows 10 1809+/Server 2019+, Ubuntu 20.04+, Debian 10+, Alpine 3.19+; 4 GB+ RAM; x64 or ARM64; internet required; Bash, Zsh, PowerShell or CMD.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/self-hosted-environments
Locator: How self-hosted environments work; Availability and limitations
Finding: Public beta on Team and Enterprise (off by default); runners on customer infrastructure execute cloud sessions; unavailable with ZDR or HIPAA; model inference still via Anthropic API unless Bedrock/Agent Platform configured; session content goes to Anthropic.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/routines
Locator: Triggers; Add an API trigger; GitHub event; Usage and limits
Finding: Cloud routines are now named cloud scheduled tasks; shared web/Desktop/CLI account records, schedule/API/GitHub triggers, research preview fire endpoint, and per-account execution. /schedule creates them and /routines remains an alias.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/agents
Locator: Five ways to work on several tasks; Choose an approach; Check on running work
Finding: Subagents, agent view (research preview), agent teams (experimental), dynamic workflows, and Projects. In every approach the workers are Claude sessions; other tools are exposed via MCP. Worktrees isolate parallel sessions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/security
Locator: Security foundation; Cloud execution security; Report a vulnerability
Finding: References Anthropic Trust Center for SOC 2 Type 2 report and ISO 27001 certificate; documents sandbox, permissions, cloud VM isolation; HackerOne vulnerability reporting.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/feature-availability
Locator: Availability by model provider; Features that require a Claude subscription
Finding: Local CLI features work across supported providers. Desktop is an exception: gateway/3P deployments can run Code on Bedrock, Google Agent Platform, Foundry, self-hosted gateways and the Claude API with Console key. Subscription-only features have separate plan restrictions.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/changelog
Locator: Version 2.1.319, November 6, 2026, and cited historical entries
Finding: Changelog reread November 7 lists v2.1.319 at the top. Historical October entries retained. No binary installed or exercised.
Retrieved: 2026-11-07

Source: https://github.com/anthropics/claude-code
Locator: Repository metadata, LICENSE.md, releases and issues
Finding: Public Claude Code repository and issue navigation reread November 7. It is a report/release repository, not proof of a product source license or reliability.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07

Source: https://status.claude.com
Locator: Summary API and incident history
Finding: Vendor status page with current component states, incident history and subscription feeds reread November 7. No uptime or response commitment inferred.
Retrieved: 2026-11-07

Source: https://status.claude.com/api/v2/summary.json
Locator: components: Claude Code; Claude Console (platform.claude.com)
Finding: Status summary read November 7 reports Claude Code and Console operational at retrieval. This is a point-in-time vendor status report, not uptime evidence.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/channels
Locator: Intro note; Research preview
Finding: Channels are in research preview, need claude.ai or Console authentication, and are unavailable on Bedrock, Agent Platform and Foundry.
Retrieved: 2026-11-07

## claude-code:OWN-07

Finding: Claude Code documentation points to the Anthropic Trust Center for compliance artifacts. The Trust Center was not fetched; SOC 2 Type 2 and ISO 27001 scope, issuer, validity and whether they cover Claude Code were not verified.

Verification: agent-cross-checked
Checked: 2026-10-09
Scheduled recheck: 2026-11-08
Plan: Not specified
Surface: Claude Code CLI
Version: Claude Code v2.1.295 (released 2026-10-08); official docs and pricing page read 2026-10-09

Vendor-stated; no certificate or report read.

Source: https://code.claude.com/docs/en/data-usage
Locator: Data training policy; Data retention; Telemetry services; Default behaviors by API provider
Finding: Consumer plans: training on opt-in setting, retention 5 years if allowed else 30 days; commercial plans no training by default, 30-day standard retention, ZDR for qualified Enterprise accounts; local transcripts kept 30 days by default; telemetry/error reporting defaults per provider with env-var opt-outs.
Retrieved: 2026-10-09

Source: https://code.claude.com/docs/en/security
Locator: Security foundation; Cloud execution security; Report a vulnerability
Finding: References Anthropic Trust Center for SOC 2 Type 2 report and ISO 27001 certificate; documents sandbox, permissions, cloud VM isolation; HackerOne vulnerability reporting.
Retrieved: 2026-10-09

## claude-code:OWN-08

Finding: Free excludes Claude Code; all paid plans include it. Evaluation requires an eligible subscription, Console/API billing or supported provider route; bundled platform credits do not pay Code usage.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Free plan excluded
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Console API credit promotions and trials were not checked.
Console API credit promotions were not checked; trial length and terms for sales-assisted Enterprise are not stated.

Source: https://code.claude.com/docs/en/authentication
Locator: Log in to Claude Code; Log in with multiple accounts
Finding: Account types: Claude Pro/Max, Teams/Enterprise, Claude Console, Bedrock/Vertex(Agent Platform)/Foundry, Claude apps gateway. Multiple accounts via separate CLAUDE_CONFIG_DIR directories. ANTHROPIC_API_KEY approval skips login.
Retrieved: 2026-11-07

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Current list prices: Pro $20 monthly or $200/year displayed as $17/month; Max $100 monthly; Ultra $200 monthly. Team 2-150 users: Standard $25 monthly/$20 annual per month, Premium $125/$100. Enterprise $20/seat/month annual, 20-seat minimum, plus API usage. All paid plans include Code; taxes excluded. Max/Ultra platform API credits cannot pay Code usage.
Retrieved: 2026-11-07

## claude-code:OWN-09

Finding: Public seat prices, plan multiples and separate API rates are stated. Enterprise has a 20-seat annual minimum; model/task costs vary, and bundled Max/Ultra platform credits exclude Code usage.

Verification: agent-cross-checked
Checked: 2026-11-07
Scheduled recheck: 2026-11-21
Plan: Not specified
Surface: Claude Code CLI; Claude Code desktop; Claude Code web/mobile; Claude Code IDE extensions
Version: First-party public docs reread 2026-11-07; changelog through v2.1.319

Judgement is the researcher classification, not a vendor statement.

Source: https://code.claude.com/docs/en/claude-code-on-the-web
Locator: Plans; GitHub authentication options; Environment expired; Limitations; Share sessions; Auto-fix pull requests
Finding: Cloud sessions on Pro, Max, Team and Enterprise (premium seats); Anthropic-managed VM by default; clone/PR creation requires GitHub (GHES for Team/Enterprise); non-GitHub repos via local bundle <100 MB; idle sessions pause and VMs are reclaimed, conversation history restored on reopen; shared rate limits, no separate compute charge; CLI --cloud and --teleport (one-way cloud to terminal); auto-fix of PR CI failures and review comments needs the Claude GitHub App.
Retrieved: 2026-11-07

Source: https://code.claude.com/docs/en/costs
Locator: Track your costs; Plan usage breakdown; Usage credits; Manage costs for your organization
Finding: /usage shows client-side session cost estimate at list price, plan usage bars and attribution to skills/subagents/plugins/MCP; usage credits past plan limits; --max-budget-usd and --max-turns in print mode; org spend limits for Team/Enterprise; average enterprise cost about $13 per developer per active day.
Retrieved: 2026-11-07

Source: https://claude.com/pricing
Locator: Individual and Team and Enterprise plan cards; feature comparison
Finding: Current list prices: Pro $20 monthly or $200/year displayed as $17/month; Max $100 monthly; Ultra $200 monthly. Team 2-150 users: Standard $25 monthly/$20 annual per month, Premium $125/$100. Enterprise $20/seat/month annual, 20-seat minimum, plus API usage. All paid plans include Code; taxes excluded. Max/Ultra platform API credits cannot pay Code usage.
Retrieved: 2026-11-07

Source: https://support.claude.com/en/articles/11145838-use-claude-code-with-your-pro-max-or-ultra-plan
Locator: Connect; usage limits; Claude Code billing
Finding: Pro, Max and Ultra include Code in a shared subscription quota; Max and Ultra platform API credits cannot pay Code usage. ANTHROPIC_API_KEY causes API billing instead of subscription. No workload usage test.
Retrieved: 2026-11-07