| name | hooks-builder |
| description | Creates Claude Code hooks executing shell commands, prompt evaluations, or multi-turn agents at 15 lifecycle events. Use when building PreToolUse validation, PostToolUse formatting, PermissionRequest auto-approval, Stop/SubagentStop evaluation, TeammateIdle prevention, TaskCompleted gating, Setup provisioning, SessionStart context injection, or deterministic agent control via blocking/non-blocking hooks. |
[H1][HOOKS-BUILDER]
Dictum: Deterministic behavior requires hooks; prompts fail execution guarantees.
Build Claude Code hooks—shell commands, prompt evaluations, or multi-turn agents execute at 15 agent lifecycle events.
Tasks:
- Read lifecycle.md — 14 events, input schemas, exit codes, blocking behavior
- Read schema.md — Configuration structure, matchers, JSON responses, hook types
- (integration) Read integration.md — Environment variables, context injection
- (scripting) Read scripting.md — Python standards, security patterns
- (recipes) Read recipes.md — Proven implementation patterns
- (troubleshooting) Read troubleshooting.md — Known issues, platform workarounds
- (prose) Load
style-standards skill — Voice, formatting, constraints
- Validate — Quality gate; see §VALIDATION
Scope:
- Event Selection: Choose hook type by automation goal (blocking vs observing).
- Configuration: Author settings.json entries with matchers and timeouts.
- Response Handling: Control agent via exit codes, JSON responses, or prompt evaluation.
References:
[1][EVENT_SELECTION]
Dictum: Automation goal determines hook type; blocking capability varies by event.
Decision Gate:
- Intercept before execution? → PreToolUse (validate/block/modify parameters)
- Control permission dialogs? → PermissionRequest (auto-approve/deny)
- React after completion? → PostToolUse (format, lint, add context)
- React after failure? → PostToolUseFailure (error handling, retry logic)
- Inject at session boundaries? → SessionStart (context), UserPromptSubmit (per-message)
- One-time provisioning? → Setup (tool installation, dependency setup via
claude --init)
- Evaluate task completion? → Stop/SubagentStop (prompt/agent type for LLM judgment)
- Coordinate teams? → TeammateIdle (prevent idle), TaskCompleted (validate completion)
- Observe subagent lifecycle? → SubagentStart/SubagentStop (logging)
Blocking Events (exit 2 blocks action):
| [INDEX] | [EVENT] | [EXIT_2_EFFECT] |
|---|
| [1] | PreToolUse | Blocks tool call; stderr shown to Claude |
| [2] | PermissionRequest | Denies the permission |
| [3] | UserPromptSubmit | Blocks prompt processing; erases prompt from context |
| [4] | Stop | Prevents Claude from stopping; continues conversation |
| [5] | SubagentStop | Prevents subagent from stopping |
| [6] | TeammateIdle | Prevents teammate from going idle; stderr = feedback |
| [7] | TaskCompleted | Prevents task completion; stderr = feedback to model |
Non-blocking Events (exit 2 shows stderr only):
| [INDEX] | [EVENT] | [EXIT_2_EFFECT] |
|---|
| [1] | PostToolUse | Shows stderr to Claude (tool already ran) |
| [2] | PostToolUseFailure | Shows stderr to Claude (tool already failed) |
| [3] | SessionStart | Shows stderr to user only |
| [4] | Setup | Shows stderr to user only |
| [6] | SessionEnd | Shows stderr to user only |
| [7] | Notification | Shows stderr to user only |
| [8] | SubagentStart | Shows stderr to user only |
| [9] | PreCompact | Shows stderr to user only |
[2][CONFIGURATION]
Dictum: Centralized configuration enables scope-aware hook precedence.
| [INDEX] | [SCOPE] | [PATH] | [USE] | [GIT] |
|---|
| [1] | User | ~/.claude/settings.json | Global, all projects | N/A |
| [2] | Project | .claude/settings.json | Shared, committed | Commit |
| [3] | Local | .claude/settings.local.json | Personal, testing | Ignore |
Precedence: Local > Project > User. Same-event hooks from all scopes run in parallel.
Snapshot: Hooks captured at startup; mid-session edits require /hooks review to reload.
[3][IMPLEMENTATION]
Dictum: Deterministic and evaluative patterns require distinct execution modes.
| [INDEX] | [TYPE] | [USE_CASE] | [TIMEOUT] | [CHARACTERISTICS] |
|---|
| [1] | command | Validation, formatting, rules | 600s | Deterministic, shell scripts |
| [2] | prompt | Complex evaluation, LLM judgment | 30s | Single-turn, context-aware |
| [3] | agent | Tool-using evaluation | 60s | Multi-turn, up to 50 turns |
Prompt/Agent Eligible Events: PreToolUse, PostToolUse, PostToolUseFailure, PermissionRequest, UserPromptSubmit, Stop, SubagentStop, TaskCompleted.
[CRITICAL] TeammateIdle does NOT support prompt or agent hooks — exit codes only.
Prompt/Agent Response Schema:
{"ok": true}
{"ok": false, "reason": "Explanation shown to Claude"}
ok: true allows the action. ok: false blocks it with the provided reason.
Command Hook Fields:
| [INDEX] | [FIELD] | [TYPE] | [DEFAULT] | [EFFECT] |
|---|
| [1] | type | string | — | "command", "prompt", or "agent" |
| [2] | command | string | — | Shell command or script path |
| [3] | timeout | number | type-specific | Seconds: command=600, prompt=30, agent=60 |
| [4] | async | boolean | false | Background execution; non-blocking |
| [5] | statusMessage | string | — | Custom spinner text during execution |
| [6] | once | boolean | false | Run once per session (skills only) |
Prompt/Agent Hook Fields:
| [INDEX] | [FIELD] | [TYPE] | [DEFAULT] | [EFFECT] |
|---|
| [1] | prompt | string | — | Instructions for LLM; $ARGUMENTS = hook JSON |
| [2] | model | string | fast model | Model to use for evaluation |
[4][SCRIPTING]
Dictum: Hook reliability requires functional pipeline patterns.
Python 3.14+ with strict typing. Zero imperative patterns.
[5][VALIDATION]
Dictum: Gates prevent incomplete artifacts.
[VERIFY] Completion:
[REFERENCE] Operational checklist: ->validation.md