agentboot 0.1.0

package/docs/claude-code-reference/feature-inventory.md

@@ -0,0 +1,906 @@
+# Claude Code Feature Inventory
+
+Complete reference of every Claude Code feature, configuration option, and capability.
+Last verified: March 2026.
+
+---
+
+## 1. CLAUDE.md System
+
+### File Locations & Precedence (highest to lowest)
+
+| Priority | Location | Scope | Shared | Excludable |
+|----------|----------|-------|--------|------------|
+| 1 | Managed policy paths (see below) | Organization | IT-deployed | No |
+| 2 | `./CLAUDE.md` or `./.claude/CLAUDE.md` | Project | Yes (git) | Yes |
+| 3 | Parent directory CLAUDE.md files (walk up tree) | Ancestors | Varies | Yes |
+| 4 | `~/.claude/CLAUDE.md` | User | No | Yes |
+| 5 | Subdirectory CLAUDE.md files | Lazy-loaded | Varies | Yes |
+
+Managed policy paths:
+- macOS: `/Library/Application Support/ClaudeCode/CLAUDE.md`
+- Linux/WSL: `/etc/claude-code/CLAUDE.md`
+- Windows: `C:\Program Files\ClaudeCode\CLAUDE.md`
+
+### @import Syntax
+
+```markdown
+@path/to/file.md           # Relative to importing file
+@./docs/api.md              # Explicit relative
+@../shared/standards.md     # Parent traversal
+@~/preferences.md           # Home directory
+@/absolute/path/file.md     # Absolute path
+```
+
+- Max recursion depth: **5 levels**
+- First external import triggers approval dialog; can decline permanently
+- Imported files expand inline at load time
+- Relative paths resolve from the **importing file's location**, not cwd
+
+### Behavioral Notes
+
+- Target **under 200 lines** per file (longer = higher context cost, lower adherence)
+- Specific, concrete instructions outperform vague goals
+- Full CLAUDE.md content **survives compaction** — re-injected fresh after `/compact`
+- Subdirectory CLAUDE.md files are **lazy-loaded** when Claude reads files in that directory
+
+### claudeMdExcludes
+
+```json
+{
+  "claudeMdExcludes": [
+    "**/monorepo/CLAUDE.md",
+    "/home/user/monorepo/other-team/.claude/rules/**"
+  ]
+}
+```
+
+- Glob patterns matched against absolute file paths
+- Configurable at any settings layer
+- **Cannot exclude managed policy CLAUDE.md**
+
+---
+
+## 2. Settings System
+
+### Configuration Scopes (highest to lowest precedence)
+
+| Scope | Location | Shared | Override |
+|-------|----------|--------|---------|
+| Managed | System paths (see §1) | IT-deployed | Cannot be overridden |
+| CLI flags | Session | No | Session only |
+| Local project | `.claude/settings.local.json` | No (gitignored) | Per-machine |
+| Shared project | `.claude/settings.json` | Yes (git) | Per-project |
+| User | `~/.claude/settings.json` | No | All projects |
+
+### Complete Settings Fields
+
+```jsonc
+{
+  // Model & behavior
+  "defaultModel": "sonnet|opus|haiku|inherit|<full-model-id>",
+  "model": "claude-opus-4-6",
+  "agent": "agent-name",                    // Set main agent for session
+  "effort": "low|medium|high|max",
+  "defaultMode": "default|acceptEdits|dontAsk|bypassPermissions|plan",
+  "skipDangerousModePermissionPrompt": true,
+
+  // Permissions
+  "permissions": {
+    "allow": ["Bash(npm run *)", "Read", "WebFetch(domain:example.com)"],
+    "ask": ["Edit"],
+    "deny": ["Bash(git push *)", "Agent(Explore)"]
+  },
+
+  // Memory
+  "autoMemoryEnabled": true,
+  "autoMemoryDirectory": "~/custom-memory-dir",
+  "claudeMdExcludes": ["pattern"],
+
+  // Environment
+  "env": {
+    "DEBUG": "true",
+    "NODE_ENV": "development"
+  },
+
+  // Working directories
+  "additionalDirectories": ["../shared", "/absolute/path"],
+
+  // Hooks
+  "disableAllHooks": false,
+  "hooks": { /* see §6 */ },
+
+  // Plugins
+  "enabledPlugins": ["plugin-name"],
+  "extraKnownMarketplaces": ["https://github.com/user/marketplace"],
+  "strictKnownMarketplaces": false,
+
+  // File suggestions
+  "autoIncludeFilesInPrompts": true,
+
+  // Sandboxing
+  "sandbox": {
+    "enabled": false,
+    "mode": "strict|moderate|permissive",
+    "readPaths": ["/tmp", "~/.ssh/**"],
+    "writePaths": ["/tmp"],
+    "allowedDomains": ["example.com", "*.trusted.com"]
+  },
+
+  // Attribution
+  "contributionAttribution": {
+    "userEmail": "user@example.com"
+  },
+
+  // Managed-only settings
+  "disableBypassPermissionsMode": "disable",
+  "allowManagedPermissionRulesOnly": false,
+  "allowManagedHooksOnly": false,
+  "allowManagedMcpServersOnly": false,
+  "blockedMarketplaces": ["untrusted-source"],
+  "allow_remote_sessions": true,
+  "forceLoginMethod": "oauth|sso|email",
+  "forceLoginOrgUUID": "uuid-here"
+}
+```
+
+### Permission Rule Syntax
+
+**Bash:**
+- Exact: `Bash(npm run build)`
+- Wildcards: `Bash(npm run *)`, `Bash(git * main)`
+- Space before `*` enforces word boundary: `Bash(ls *)` matches `ls -la` but not `lsof`
+- Shell operator aware: `Bash(safe *)` won't allow `safe && dangerous`
+
+**Read/Edit path patterns:**
+- Project-relative: `Edit(/src/**/*.ts)` → `<project>/src/**/*.ts`
+- Home: `Read(~/.zshrc)`
+- Absolute: `Edit(//tmp/file.txt)` → `/tmp/file.txt`
+- CWD-relative: `Read(src/*.js)` or `Read(./src/*.js)`
+- Gitignore glob patterns: `*` (single dir), `**` (recursive)
+
+**Other tools:**
+- WebFetch: `WebFetch(domain:example.com)`
+- MCP: `mcp__server-name`, `mcp__server__tool-name`, `mcp__server__.*`
+- Agent: `Agent(Explore)`, `Agent(my-agent)`
+- Skill: `Skill(name)`, `Skill(name *)`
+
+**Evaluation order:** deny → ask → allow (first match wins). Deny always takes precedence.
+
+### Permission Modes
+
+| Mode | Behavior |
+|------|----------|
+| `default` | Standard prompts for each action |
+| `acceptEdits` | Auto-accept file edits for session |
+| `dontAsk` | Auto-deny unless pre-approved in allow list |
+| `plan` | Read-only — no file edits or commands |
+| `bypassPermissions` | Skip prompts (except `.git`, `.claude`, `.vscode`, `.idea`) |
+
+---
+
+## 3. Agent System
+
+### File Format: `.claude/agents/{name}/CLAUDE.md`
+
+```yaml
+---
+name: agent-id                               # Required (lowercase, hyphens, max 64 chars)
+description: When to delegate to this agent  # Required
+tools: Read, Grep, Glob, Bash               # Optional: tool allowlist
+disallowedTools: Write, Edit, Agent          # Optional: tool denylist
+model: sonnet|opus|haiku|inherit|<id>        # Optional (default: inherit)
+permissionMode: default|acceptEdits|...      # Optional
+maxTurns: 50                                 # Optional: agentic turn limit
+skills:                                      # Optional: preload skill content
+  - skill-name
+mcpServers:                                  # Optional: scoped MCP servers
+  server-name:
+    type: stdio
+    command: npx
+    args: ["-y", "package"]
+  existing-server: {}                        # Reference existing by name
+hooks:                                       # Optional: agent-specific hooks
+  PreToolUse:
+    - matcher: "Bash"
+      hooks:
+        - type: command
+          command: "./scripts/validate.sh"
+memory: user|project|local                   # Optional: persistent memory scope
+background: true|false                       # Optional: run as background task
+isolation: worktree                          # Optional: git worktree isolation
+---
+
+Agent system prompt in markdown.
+```
+
+### Storage Locations (priority order)
+
+1. `--agents` CLI flag (session only, JSON)
+2. `.claude/agents/{name}/CLAUDE.md` (project, committed)
+3. `~/.claude/agents/{name}/CLAUDE.md` (user, all projects)
+4. Plugin's `agents/` directory
+
+### Built-in Agents
+
+| Agent | Purpose | Default Tools |
+|-------|---------|---------------|
+| `Explore` | Fast codebase exploration | Read, Glob, Grep, Bash, LSP |
+| `Plan` | Architecture and planning | Read, Glob, Grep, Bash, LSP, WebFetch, WebSearch |
+| `general-purpose` | Multi-step tasks | All tools except Agent |
+| `Bash` | Shell command execution | Bash only |
+| `statusline-setup` | Status line config | Read, Edit |
+| `Claude Code Guide` | CC documentation Q&A | Glob, Grep, Read, WebFetch, WebSearch |
+
+### Invocation Methods
+
+1. **Automatic delegation** — Claude matches task to agent `description`
+2. **@-mention** — `@"agent-name (agent)"` guarantees delegation
+3. **CLI flag** — `claude --agent agent-name`
+4. **Settings** — `"agent": "agent-name"` in settings.json
+5. **Natural language** — name the agent; Claude may delegate
+
+### Context Behavior
+
+- Each invocation creates **fresh context** (unless resumed via SendMessage)
+- CLAUDE.md files still load in subagent context
+- Preloaded skills injected at startup
+- Subagents **cannot spawn other subagents** (no nesting)
+- Main conversation compaction doesn't affect subagent transcripts
+- Subagent transcripts: `~/.claude/projects/{project}/{sessionId}/subagents/agent-{agentId}.jsonl`
+
+### Memory Scope
+
+| Scope | Directory | Shared |
+|-------|-----------|--------|
+| `user` | `~/.claude/agent-memory/{name}/` | No (cross-project) |
+| `project` | `.claude/agent-memory/{name}/` | Yes (git) |
+| `local` | `.claude/agent-memory-local/{name}/` | No (gitignored) |
+
+First 200 lines of agent's `MEMORY.md` loaded at startup. Read/Write/Edit tools auto-enabled when memory is set.
+
+### Background Agents
+
+- Set `background: true` in frontmatter or press `Ctrl+B` while running
+- Prompt for permissions upfront; auto-deny unlisted tools
+- Run concurrently with main session
+- `AskUserQuestion` calls fail silently; agent continues
+
+### Worktree Isolation
+
+- Set `isolation: worktree` in frontmatter
+- Agent gets isolated copy of repo (git worktree)
+- Auto-cleaned if agent makes no changes
+- If changes made, worktree path and branch returned
+
+---
+
+## 4. Skills System
+
+### File Format: `.claude/skills/{name}/SKILL.md`
+
+```yaml
+---
+name: skill-name                             # Optional (defaults to directory name)
+description: When to use this skill          # Recommended (auto-invoke trigger)
+argument-hint: "[version] [environment]"     # Optional: CLI autocomplete hint
+disable-model-invocation: false              # Optional: user-only trigger
+user-invocable: true                         # Optional: hide from / menu
+allowed-tools: Read, Grep, Bash              # Optional: tool allowlist
+model: sonnet|opus|haiku|inherit             # Optional: override model
+context: fork                                # Optional: run in isolated subagent
+agent: Explore|Plan|general-purpose|{name}   # Optional: subagent type (requires context: fork)
+hooks:                                       # Optional: skill-specific hooks
+  PreToolUse:
+    - matcher: "Bash"
+      hooks: [...]
+---
+
+Skill instructions.
+```
+
+### String Substitutions
+
+| Placeholder | Expands To |
+|-------------|-----------|
+| `$ARGUMENTS` | All arguments passed to skill |
+| `$ARGUMENTS[N]` or `$N` | Specific argument by index (0-based) |
+| `${CLAUDE_SESSION_ID}` | Current session ID |
+| `${CLAUDE_SKILL_DIR}` | Directory containing SKILL.md |
+
+### Dynamic Context Injection
+
+Syntax: `` !`command` `` (backticks around shell command)
+
+- Executed **before** Claude sees skill content
+- Output replaces the placeholder inline
+- Claude sees the result, not the command
+- Example: `` !`gh pr diff` `` fetches live PR data at invocation time
+
+### Supporting Files
+
+```
+.claude/skills/my-skill/
+├── SKILL.md              # Required
+├── reference.md          # Optional: detailed docs (on-demand)
+├── examples.md           # Optional: usage examples
+└── scripts/
+    └── helper.py         # Optional: executable scripts
+```
+
+### Storage Locations
+
+1. Enterprise: Managed settings (all users)
+2. Personal: `~/.claude/skills/{name}/SKILL.md`
+3. Project: `.claude/skills/{name}/SKILL.md`
+4. Plugin: `{plugin}/skills/{name}/SKILL.md`
+
+### Invocation Control
+
+| Config | User can `/invoke` | Claude auto-invokes |
+|--------|-------------------|-------------------|
+| Default | Yes | Yes (via description match) |
+| `disable-model-invocation: true` | Yes | No |
+| `user-invocable: false` | No | Yes |
+
+### Bundled Skills
+
+| Skill | Trigger |
+|-------|---------|
+| `/batch <instruction>` | Large-scale parallel changes (5-30 worktrees) |
+| `/claude-api` | Auto-triggered with anthropic/SDK imports |
+| `/debug [description]` | Troubleshoot via debug log |
+| `/loop [interval] <prompt>` | Recurring prompts |
+| `/simplify [focus]` | Review changes with 3 parallel agents |
+
+---
+
+## 5. Rules System
+
+### File Format: `.claude/rules/{topic}.md`
+
+```yaml
+---
+paths:                              # Optional glob patterns
+  - "src/api/**/*.ts"
+  - "src/**/*.{ts,tsx}"
+  - "*.md"
+---
+
+Rule content in markdown.
+```
+
+### Behavior
+
+- Rules **without** `paths:` — loaded at session start (always active)
+- Rules **with** `paths:` — loaded on-demand when Claude reads matching files
+- Supports brace expansion: `*.{ts,tsx,js}`
+- Supports `**` recursive glob
+
+### Storage & Priority
+
+| Location | Scope | Priority |
+|----------|-------|----------|
+| `.claude/rules/` | Project | Higher |
+| `~/.claude/rules/` | User (all projects) | Lower |
+
+- Symlinks supported for sharing rules across projects
+- Subdirectory `.claude/rules/` auto-discovered recursively
+
+---
+
+## 6. Hooks System (25 Events)
+
+### All Hook Events
+
+| Event | Matcher | Fires When | Exit 2 Blocks |
+|-------|---------|------------|---------------|
+| `SessionStart` | startup, resume, clear, compact | Session begins/resumes | No |
+| `SessionEnd` | clear, logout, prompt_input_exit, bypass_permissions_disabled, other | Session terminates | No |
+| `InstructionsLoaded` | session_start, nested_traversal, path_glob_match, include, compact | CLAUDE.md/rules loaded | No |
+| `UserPromptSubmit` | (none) | Before processing user prompt | **Yes** |
+| `PreToolUse` | Tool name | Before tool execution | **Yes** |
+| `PostToolUse` | Tool name | After tool succeeds | No |
+| `PostToolUseFailure` | Tool name | After tool fails | No |
+| `PermissionRequest` | Tool name | Permission dialog appears | **Yes** |
+| `Stop` | (none) | Claude finishes response | **Yes** |
+| `StopFailure` | rate_limit, authentication_failed, billing_error, invalid_request, server_error, max_output_tokens, unknown | Turn ends with API error | No |
+| `SubagentStart` | Agent type name | Subagent spawned | No |
+| `SubagentStop` | Agent type name | Subagent finishes | **Yes** |
+| `Notification` | permission_prompt, idle_prompt, auth_success, elicitation_dialog | Notification sent | No |
+| `TeammateIdle` | (none) | Agent team teammate idle | No |
+| `TaskCompleted` | (none) | Task marked complete | No |
+| `ConfigChange` | user_settings, project_settings, local_settings, policy_settings, skills | Config file changes | **Yes** |
+| `PreCompact` | manual, auto | Before context compaction | No |
+| `PostCompact` | manual, auto | After compaction completes | No |
+| `WorktreeCreate` | (none) | Worktree creation initiated | **Yes** |
+| `WorktreeRemove` | (none) | Worktree removal | No |
+| `Elicitation` | MCP server name | MCP requests user input | **Yes** |
+| `ElicitationResult` | MCP server name | User responds to elicitation | **Yes** |
+
+### Hook Types
+
+**Command hook:**
+```json
+{
+  "type": "command",
+  "command": "script.sh",
+  "timeout": 600,
+  "async": false,
+  "statusMessage": "Running validation..."
+}
+```
+
+**HTTP hook:**
+```json
+{
+  "type": "http",
+  "url": "http://localhost:8080/hooks",
+  "headers": {"Authorization": "Bearer $TOKEN"},
+  "allowedEnvVars": ["TOKEN"],
+  "timeout": 30
+}
+```
+
+**Prompt hook (LLM evaluation):**
+```json
+{
+  "type": "prompt",
+  "prompt": "Evaluate whether this is safe: $ARGUMENTS",
+  "model": "haiku",
+  "timeout": 30
+}
+```
+
+**Agent hook (subagent with tools):**
+```json
+{
+  "type": "agent",
+  "prompt": "Analyze: $ARGUMENTS",
+  "model": "inherit",
+  "timeout": 60
+}
+```
+
+### Hook Input (stdin JSON)
+
+```json
+{
+  "session_id": "uuid",
+  "transcript_path": "/path/to/transcript.jsonl",
+  "cwd": "/current/directory",
+  "permission_mode": "default",
+  "hook_event_name": "PreToolUse",
+  "agent_id": "subagent-id",
+  "agent_type": "agent-name"
+}
+```
+
+Additional fields vary by event (tool_name, tool_input for PreToolUse; response for Stop; etc.).
+
+### Hook Output (stdout JSON)
+
+```json
+{
+  "continue": true,
+  "stopReason": "message if continue=false",
+  "suppressOutput": false,
+  "systemMessage": "injected warning",
+  "decision": "block|allow|deny|ask",
+  "reason": "explanation",
+  "hookSpecificOutput": { }
+}
+```
+
+### Exit Codes
+
+| Code | Meaning |
+|------|---------|
+| 0 | Success — process JSON output |
+| 2 | Block action — show stderr to user |
+| Other | Non-blocking error — show in verbose mode |
+
+### Matcher Patterns for MCP Tools
+
+- `mcp__<server>__<tool>` — exact match
+- `mcp__memory__.*` — all tools from memory server
+- `mcp__.*__write.*` — write operations from any server
+
+### Configuration Locations
+
+- `~/.claude/settings.json` (user)
+- `.claude/settings.json` (project, shared)
+- `.claude/settings.local.json` (project, local)
+- Managed policy settings
+- Plugin `hooks/hooks.json`
+- Agent/Skill frontmatter `hooks:` field
+
+---
+
+## 7. MCP System
+
+### Configuration: `.mcp.json`
+
+```json
+{
+  "mcpServers": {
+    "server-name": {
+      "type": "http|sse|stdio|ws",
+      "url": "https://example.com/mcp",
+      "command": "/path/to/server",
+      "args": ["--arg1", "value"],
+      "env": {"KEY": "${ENV_VAR:-default}"},
+      "headers": {"Authorization": "Bearer ${TOKEN}"},
+      "oauth": {
+        "clientId": "id",
+        "clientSecret": "secret",
+        "callbackPort": 8080,
+        "authServerMetadataUrl": "https://auth.example.com/.well-known"
+      }
+    }
+  }
+}
+```
+
+### Transport Types
+
+| Type | Use Case |
+|------|----------|
+| `http` | Remote HTTP server (recommended) |
+| `stdio` | Local process (npx, python, etc.) |
+| `ws` | WebSocket |
+| `sse` | Server-Sent Events (deprecated) |
+
+### Installation Scopes
+
+| Scope | Location | CLI Flag |
+|-------|----------|----------|
+| Local (default) | `~/.claude.json` | `--scope local` |
+| Project | `.mcp.json` | `--scope project` |
+| User | `~/.claude.json` | `--scope user` |
+| Managed | System paths | IT-deployed |
+
+### CLI Commands
+
+```bash
+claude mcp add --transport http name https://url
+claude mcp add --transport stdio name -- npx package
+claude mcp add-json name '{"type":"http","url":"..."}'
+claude mcp add-from-claude-desktop
+claude mcp list
+claude mcp get name
+claude mcp remove name
+claude mcp reset-project-choices
+claude mcp serve                    # Claude as MCP server
+```
+
+### Managed MCP
+
+**Option 1: Exclusive control** — `managed-mcp.json` at system paths. Users cannot add/modify.
+
+**Option 2: Policy-based allowlist/denylist:**
+```json
+{
+  "allowedMcpServers": [
+    { "serverName": "github" },
+    { "serverCommand": ["npx", "-y", "package"] },
+    { "serverUrl": "https://mcp.company.com/*" }
+  ],
+  "deniedMcpServers": [
+    { "serverName": "untrusted" }
+  ]
+}
+```
+Denylist takes precedence. `managed-mcp.json` overrides both.
+
+### MCP Resources
+
+- Reference: `@server:protocol://path`
+- Auto-fetched as attachments
+- Fuzzy-searchable in `@` autocomplete
+- Example: `@github:issue://123`
+
+### MCP Prompts as Commands
+
+- Format: `/mcp__servername__promptname`
+- Discoverable via `/` menu
+- Arguments space-separated
+
+### MCP Output Limits
+
+- Warning threshold: 10,000 tokens
+- Default max: 25,000 tokens
+- Override: `MAX_MCP_OUTPUT_TOKENS=50000 claude`
+
+### Tool Search (Deferred Discovery)
+
+- Auto-enabled when tool descriptions exceed 10% of context
+- `ENABLE_TOOL_SEARCH=auto:5` (custom threshold percentage)
+- `ENABLE_TOOL_SEARCH=false` to disable
+
+---
+
+## 8. Memory System
+
+### Directory Structure
+
+```
+~/.claude/projects/<project>/memory/
+├── MEMORY.md          # Index — first 200 lines loaded at session start
+├── topic-a.md         # Topic files — loaded on-demand
+└── topic-b.md
+```
+
+- `<project>` derived from git repo name
+- All worktrees in same repo share one memory directory
+- Machine-local (not synced)
+
+### Settings
+
+| Setting | Type | Default | Scope |
+|---------|------|---------|-------|
+| `autoMemoryEnabled` | boolean | true | Any |
+| `autoMemoryDirectory` | string | (auto) | User/local/managed only |
+| `CLAUDE_CODE_DISABLE_AUTO_MEMORY` | env var | unset | Session |
+
+---
+
+## 9. CLI Features
+
+### Key Flags
+
+```bash
+# Session
+claude                                  # Interactive mode
+claude "prompt"                         # Start with prompt
+claude -p "prompt"                      # Print mode (non-interactive)
+claude -c                               # Continue last session
+claude -r "session-id|name"             # Resume specific session
+claude --resume id --fork-session       # Fork from existing session
+
+# Model & effort
+claude --model opus|sonnet|haiku
+claude --effort low|medium|high|max
+claude --fallback-model sonnet          # Print mode only
+
+# Permissions
+claude --permission-mode default|acceptEdits|plan|dontAsk|bypassPermissions
+claude --tools "Bash,Edit,Read"
+claude --allowedTools "Bash(npm *)"
+claude --disallowedTools "Bash(git push *)"
+
+# System prompt
+claude --system-prompt "text"
+claude --system-prompt-file ./file.txt
+claude --append-system-prompt "extra"
+claude --append-system-prompt-file ./extra.txt
+
+# Agents
+claude --agent agent-name
+claude --agents '{"name": {...}}'
+
+# Worktrees
+claude --worktree feature-name
+claude -w feature-name
+
+# Working dirs
+claude --add-dir ../lib ../apps
+
+# MCP
+claude --mcp-config ./mcp.json
+claude --strict-mcp-config
+claude --chrome / --no-chrome
+
+# Output (print mode)
+claude --output-format text|json|stream-json
+claude --input-format text|stream-json
+claude --json-schema '{...}'
+claude --max-turns 3
+claude --max-budget-usd 5.00
+claude --no-session-persistence
+
+# Advanced
+claude --debug "api,hooks"
+claude --verbose
+claude --betas interleaved-thinking
+claude --init / --init-only / --maintenance
+claude --from-pr 123
+claude --remote "task"
+claude --teleport
+claude --name "session-name"
+claude --session-id "uuid"
+claude --ide
+```
+
+---
+
+## 10. Built-in Tools (35)
+
+| Tool | Permission | Purpose |
+|------|-----------|---------|
+| Agent | No | Spawn subagent |
+| AskUserQuestion | No | Multiple-choice questions to user |
+| Bash | Yes | Execute shell commands |
+| CronCreate | No | Schedule recurring/one-shot prompt |
+| CronDelete | No | Cancel scheduled task |
+| CronList | No | List scheduled tasks |
+| Edit | Yes | Targeted file edits (string replacement) |
+| EnterPlanMode | No | Switch to plan mode |
+| EnterWorktree | No | Create git worktree |
+| ExitPlanMode | Yes | Exit plan mode |
+| ExitWorktree | No | Exit worktree |
+| Glob | No | File pattern matching |
+| Grep | No | Pattern search in files |
+| ListMcpResourcesTool | No | List MCP server resources |
+| LSP | No | Language server (type errors, navigation) |
+| NotebookEdit | Yes | Modify Jupyter notebook cells |
+| Read | No | Read file contents (text, images, PDFs, notebooks) |
+| ReadMcpResourceTool | No | Read specific MCP resource |
+| Skill | Yes | Execute skill |
+| TaskCreate | No | Create task |
+| TaskGet | No | Get task details |
+| TaskList | No | List all tasks |
+| TaskOutput | No | Retrieve background task output |
+| TaskStop | No | Kill background task |
+| TaskUpdate | No | Update task status/deps |
+| TodoWrite | No | Session task checklist |
+| ToolSearch | No | Search for deferred tools |
+| WebFetch | Yes | Fetch URL content |
+| WebSearch | Yes | Web search |
+| Write | Yes | Create/overwrite files |
+
+---
+
+## 11. Interactive Features
+
+### Keyboard Shortcuts
+
+| Shortcut | Action |
+|----------|--------|
+| `Ctrl+C` | Cancel current input/generation |
+| `Ctrl+D` | Exit session |
+| `Ctrl+F` | Kill all background agents (press twice) |
+| `Ctrl+L` | Clear screen |
+| `Ctrl+O` | Toggle verbose output |
+| `Ctrl+R` | Reverse history search |
+| `Ctrl+G` | Open in text editor |
+| `Ctrl+V` / `Cmd+V` | Paste image from clipboard |
+| `Ctrl+B` | Background running tasks |
+| `Ctrl+T` | Toggle task list |
+| `Esc+Esc` | Rewind or summarize |
+| `Shift+Tab` / `Alt+M` | Toggle permission modes |
+| `Alt+P` / `Cmd+P` | Switch model |
+| `Alt+T` / `Cmd+T` | Toggle extended thinking |
+| `Ctrl+K` | Delete to end of line |
+| `Ctrl+U` | Delete entire line |
+| `Ctrl+Y` | Paste deleted text |
+| `Alt+B` / `Alt+F` | Word navigation |
+
+### Input Modes
+
+- `\` + `Enter` — newline (all terminals)
+- `Alt+Enter` — newline (macOS default)
+- `Shift+Enter` — newline (iTerm2, WezTerm, Ghostty, Kitty)
+- `!command` — direct Bash execution
+- `@path` — file mention
+- `/` — skill/command autocomplete
+- `Space` (held) — push-to-talk dictation
+
+### Vim Mode
+
+Enable with `/vim`. Supports: `h/j/k/l`, `w/e/b`, `0/$`, `gg/G`, `f/F/t/T`, `dd/D/dw`, `cc/C/cw`, `yy/yw`, `p/P`, `>>/<<`, `.`, text objects (`iw/aw`, `i"/a"`, `i(/a(`, `i[/a[`, `i{/a{`).
+
+### Keybindings Customization
+
+File: `~/.claude/keybindings.json`
+
+```json
+{
+  "bindings": [
+    {
+      "context": "Chat",
+      "bindings": {
+        "ctrl+e": "chat:externalEditor",
+        "ctrl+u": null
+      }
+    }
+  ]
+}
+```
+
+Contexts: Global, Chat, Autocomplete, Settings, Confirmation, Tabs, Help, Transcript, HistorySearch, Task, ThemePicker, Attachments, Footer, MessageSelector, DiffDialog, ModelPicker, Select, Plugin.
+
+Chords supported: `"ctrl+k ctrl+s": "action"`.
+
+Reserved (cannot rebind): `Ctrl+C`, `Ctrl+D`.
+
+---
+
+## 12. Agent Teams & Task System
+
+### Task Tools
+
+- `TaskCreate` — create task with description, dependencies
+- `TaskGet` — get task by ID
+- `TaskList` — list all tasks
+- `TaskUpdate` — update status (pending → in_progress → completed), add dependencies
+- `TaskOutput` — retrieve background task output
+- `TaskStop` — kill running background task
+
+### Multi-Agent Coordination
+
+- `TeammateIdle` hook — fires when an agent team teammate is idle
+- `TaskCompleted` hook — fires when a task is marked complete
+- `--teammate-mode auto|in-process|tmux` — how agents coordinate
+
+### Cron (Scheduled Tasks)
+
+- `CronCreate` — schedule recurring or one-shot prompts
+- `CronDelete` — cancel scheduled task
+- `CronList` — list all scheduled tasks
+
+---
+
+## 13. Git & Worktree Features
+
+### Worktrees
+
+- CLI: `claude --worktree feature-name` or `-w`
+- Tools: `EnterWorktree` / `ExitWorktree`
+- Agent: `isolation: worktree` in frontmatter
+- Location: `<repo>/.claude/worktrees/<name>`
+- Hooks: `WorktreeCreate`, `WorktreeRemove`
+- Auto-cleanup if subagent makes no changes
+
+### Session-PR Linking
+
+- Auto-linked when PR created via `gh pr create` in session
+- Resume: `claude --from-pr 123` or `--from-pr <url>`
+
+---
+
+## 14. Environment Variables
+
+| Variable | Purpose |
+|----------|---------|
+| `CLAUDE_ENV_FILE` | Persist env vars across Bash commands |
+| `CLAUDE_CODE_REMOTE` | "true" if cloud session |
+| `CLAUDE_PROJECT_DIR` | Project root path |
+| `CLAUDE_PLUGIN_ROOT` | Plugin installation directory |
+| `CLAUDE_PLUGIN_DATA` | Plugin persistent data directory |
+| `CLAUDE_CODE_DISABLE_AUTO_MEMORY` | Disable auto memory |
+| `CLAUDE_CODE_TASK_LIST_ID` | Named task list |
+| `CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION` | Enable/disable prompt suggestions |
+| `CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR` | Keep project cwd in Bash |
+| `MAX_MCP_OUTPUT_TOKENS` | Override MCP output limit |
+| `ENABLE_TOOL_SEARCH` | Control deferred tool discovery |
+
+---
+
+## 15. Configuration File Paths (Complete)
+
+| Path | Purpose |
+|------|---------|
+| `~/.claude/` | User config home |
+| `~/.claude/settings.json` | User settings |
+| `~/.claude/keybindings.json` | Keyboard shortcuts |
+| `~/.claude/CLAUDE.md` | User-level instructions |
+| `~/.claude/agents/` | Personal agents |
+| `~/.claude/skills/` | Personal skills |
+| `~/.claude/rules/` | Personal rules |
+| `~/.claude/projects/{project}/memory/` | Auto memory |
+| `~/.claude.json` | MCP config (local/user scope) |
+| `.claude/settings.json` | Project settings (shared) |
+| `.claude/settings.local.json` | Project settings (local) |
+| `.claude/CLAUDE.md` | Project instructions |
+| `.claude/agents/` | Project agents |
+| `.claude/skills/` | Project skills |
+| `.claude/rules/` | Project rules |
+| `.claude/agent-memory/` | Agent memory (shared) |
+| `.claude/agent-memory-local/` | Agent memory (local) |
+| `.claude/worktrees/` | Worktree isolation |
+| `.mcp.json` | MCP config (project scope) |
+| `CLAUDE.md` | Project instructions (root) |
+| `/Library/Application Support/ClaudeCode/` | Managed policy (macOS) |
+| `/etc/claude-code/` | Managed policy (Linux/WSL) |
+| `C:\Program Files\ClaudeCode\` | Managed policy (Windows) |