hook-creator（钩子创建者）

Hook Creator Skill

Creates, validates, and registers hooks for the multi-agent orchestration framework.

ROUTER UPDATE REQUIRED (CRITICAL - DO NOT SKIP)

After creating ANY hook, you MUST update documentation:

1. Add to .claude/hooks/README.md under appropriate category
2. Register in config.yaml or settings.json if required
3. Update learnings.md with hook summary

Verification:

grep "<hook-name>" .claude/hooks/README.md || echo "ERROR: hooks/README.md NOT UPDATED!"

WHY: Hooks not documented are invisible and unmaintainable.

Overview

This skill creates hooks for the Claude Code framework:

• Pre-tool execution - Safety validation before commands run
• Post-tool execution - Logging, memory updates, telemetry
• Session lifecycle - Initialize context, cleanup on exit
• Memory management - Auto-extract learnings, format memory files
• Routing enforcement - Ensure Router-First protocol compliance

Hook Types

Hook-Agent Archetype Reference

When creating hooks, determine which agent archetypes will be governed by the new hook. See .claude/docs/@HOOK_AGENT_MAP.md for:

• Section 1: Full hook-agent matrix
• Section 2: Archetype hook sets (Router, Implementer, Reviewer, Documenter, Orchestrator, Researcher)

After creating a hook, you MUST add it to both the matrix AND update affected agents' ## Enforcement Hooks sections.

Claude Code Hook Types

Workflow Steps

Step 0: Existence Check and Updater Delegation (MANDATORY - FIRST STEP)

BEFORE creating any hook file, check if it already exists:

1. Check if hook already exists: test -f.claude/hooks/<category>/<hook-name>.cjs && echo "EXISTS" || echo "NEW"
2. If hook EXISTS:

- DO NOT proceed with creation - Invoke artifact-updater workflow instead: Skill({skill: 'artifact-updater', args: '--type hook --path.claude/hooks/<category>/<hook-name>.cjs --changes "<description of requested changes>"',}); - Return updater result and STOP

1. If hook is NEW:

- Continue with Step 0.5 below

Step 0.1: Smart Duplicate Detection (MANDATORY)

Before proceeding with creation, run the 3-layer duplicate check:

const { checkDuplicate } = require('.claude/lib/creation/duplicate-detector.cjs');
const result = checkDuplicate({
  artifactType: 'hook',
  name: proposedName,
  description: proposedDescription,
  keywords: proposedKeywords || [],
});

Handle results:

• EXACT_MATCH: Stop creation. Route to hook-updater skill instead: Skill({skill: 'hook-updater'})
• REGISTRY_MATCH: Warn user — artifact is registered but file may be missing. Investigate before creating. Ask user to confirm.
• SIMILAR_FOUND: Display candidates with scores. Ask user: "Similar artifact(s) exist. Continue with new creation or update existing?"
• NO_MATCH: Proceed to Step 0.5 (companion check).

Override: If user explicitly passes --force, skip this check entirely.

Step 0.5: Companion Check

Before proceeding with creation, run the ecosystem companion check:

1. Use companion-check.cjs from .claude/lib/creators/companion-check.cjs
2. Call checkCompanions("hook", "{hook-name}") to identify companion artifacts
3. Review the companion checklist — note which required/recommended companions are missing
4. Plan to create or verify missing companions after this artifact is complete
5. Include companion findings in post-creation integration notes

This step is informational (does not block creation) but ensures the full artifact ecosystem is considered.

Reference Hook

Use .claude/lib/routing/routing-table.cjs as the canonical routing reference.

Before finalizing any hook, compare against routing-table structure:

• Has proper CommonJS exports (module.exports)
• Exports required functions for hook type (validate, main, etc.)
• Has comprehensive test file (.test.cjs)
• Has proper error handling (try/catch, graceful fallbacks)
• Returns correct response format for hook type

Step 1: Gather Hook Requirements

Before creating a hook, gather:

1. Purpose: What should this hook do?
2. Trigger: When should it run? (pre-tool, post-tool, session event)
3. Target tools: Which tools does it apply to? (Bash, Write, Edit, Read)
4. Behavior: Block operation or warn only?
5. Exit codes: What indicates success/failure?

// Example requirements gathering
{
  purpose: "Validate git push commands to prevent force push",
  trigger: "pre-tool (Bash)",
  target_tools: ["Bash"],
  behavior: "block if force push detected",
  exit_codes: { 0: "allow", 1: "block" }
}

Step 2: Determine Hook Type and Location

Naming Convention: <action>-<target>.cjs

Examples:

• validate-git-force-push.cjs
• enforce-tdd-workflow.cjs
• extract-workflow-learnings.cjs
• memory-reminder.cjs

Step 3: Generate Hook Code (CJS Format)

All hooks use CommonJS format and follow this template:

'use strict';

/**
 * {Hook Name}
 *
 * Type: {pre|post}-{tool} | session-{start|end} | user-prompt
 * Purpose: {One line description}
 * Trigger: {When this hook runs}
 *
 * Exit codes:
 * - 0: Allow operation (with optional warning)
 * - 1: Block operation (when in blocking mode)
 *
 * Environment:
 *   {HOOK_NAME}_MODE=block|warn|off (default: warn unless explicitly required)
 */

const fs = require('fs');
const path = require('path');

// Find project root by looking for .claude directory
function findProjectRoot() {
  let dir = __dirname;
  while (dir !== path.parse(dir).root) {
    if (fs.existsSync(path.join(dir, '.claude'))) {
      return dir;
    }
    dir = path.dirname(dir);
  }
  return process.cwd();
}

const PROJECT_ROOT = findProjectRoot();
const ENFORCEMENT_MODE = process.env.HOOK_NAME_MODE || 'warn';

/**
 * Parse hook input from Claude Code
 * Claude Code passes JSON via process.argv[2] for hooks
 * @returns {Object|null} Parsed hook input or null
 */
function parseHookInput() {
  try {
    if (process.argv[2]) {
      return JSON.parse(process.argv[2]);
    }
  } catch (e) {
    // Fallback for testing or invalid input
  }
  return null;
}

/**
 * Validate hook - called by Claude Code or programmatically
 * @param {Object} context - Hook context with tool info
 * @param {string} context.tool - Tool name (Bash, Write, Edit, Read)
 * @param {Object} context.parameters - Tool parameters
 * @returns {Object} Validation result with valid (boolean), error (string), and optional warning (string)
 */
function validate(context) {
  const { tool, parameters } = context;

  // YOUR VALIDATION LOGIC HERE

  // Return validation result
  return { valid: true, error: '' };
}

/**
 * Main execution for CLI hook usage
 */
function main() {
  // Skip if enforcement is off
  if (ENFORCEMENT_MODE === 'off') {
    process.exit(0);
  }

  const hookInput = parseHookInput();
  if (!hookInput) {
    process.exit(0);
  }

  // Get tool name and input
  const toolName = hookInput.tool_name || hookInput.tool;
  const toolInput = hookInput.tool_input || hookInput.input || {};

  // Run validation
  const result = validate({ tool: toolName, parameters: toolInput });

  if (!result.valid) {
    if (ENFORCEMENT_MODE === 'block') {
      console.error(`BLOCKED: ${result.error}`);
      process.exit(1);
    } else {
      console.warn(`WARNING: ${result.error}`);
      process.exit(0);
    }
  }

  if (result.warning) {
    console.warn(`WARNING: ${result.warning}`);
  }

  process.exit(0);
}

// Run main if executed directly
if (require.main === module) {
  main();
}

// Export for programmatic use and testing
module.exports = {
  validate,
  findProjectRoot,
  PROJECT_ROOT,
};

Step 4: Create Test File

Every hook MUST have a corresponding test file:

'use strict';

const { validate } = require('./hook-name.cjs');

describe('Hook Name', () => {
  test('allows valid operations', () => {
    const result = validate({
      tool: 'Bash',
      parameters: { command: 'git status' },
    });
    expect(result.valid).toBe(true);
    expect(result.error).toBe('');
  });

  test('blocks dangerous operations', () => {
    const result = validate({
      tool: 'Bash',
      parameters: { command: 'git push --force' },
    });
    expect(result.valid).toBe(false);
    expect(result.error).toContain('force push');
  });

  test('handles missing parameters gracefully', () => {
    const result = validate({
      tool: 'Bash',
      parameters: {},
    });
    expect(result.valid).toBe(true);
  });
});

Step 5: Register Hook (If Needed)

Some hooks require registration in config files:

For pre/post-tool hooks (settings.json):

{
  "hooks": {
    "pre-tool": [".claude/hooks/safety/hook-name.cjs"],
    "post-tool": [".claude/hooks/memory/hook-name.cjs"]
  }
}

For event hooks (config.yaml):

hooks:
  UserPromptSubmit:
    - path: .claude/hooks/routing/hook-name.cjs
      type: command
  SessionStart:
    - path: .claude/hooks/session/hook-name.cjs
      type: command

Step 6: Update Documentation (MANDATORY - BLOCKING)

After creating a hook, update .claude/hooks/README.md:

#### {Hook Name} (`hook-name.cjs`)

{Description of what the hook does}

**When it runs:** {Trigger condition}
**What it checks/does:** {Detailed behavior}

Verify with:

grep "hook-name" .claude/hooks/README.md || echo "ERROR: Not documented!"

Step 7: System Impact Analysis (MANDATORY)

This analysis is MANDATORY. Hook creation is INCOMPLETE without it.

After creating a hook:

1. Settings Registration (BLOCKING)

- Add to.claude/settings.json PreToolUse/PostToolUse/etc. - Verify with: grep "hook-name".claude/settings.json

1. Test Coverage (BLOCKING)

- Create.test.cjs file with minimum 10 test cases - Run tests: node.claude/hooks/<category>/<name>.test.cjs

1. Documentation

- Update.claude/docs/ if hook adds new capability - Add usage examples

1. Related Hooks

- Check if hook affects other hooks in same trigger category - Document interaction patterns

Full Checklist:

[HOOK-CREATOR] System Impact Analysis for: <hook-name>

1. HOOK FILE CREATED
   [ ] Created at .claude/hooks/<type>/<hook-name>.cjs
   [ ] Follows CJS format with validate() export
   [ ] Has main() function for CLI execution
   [ ] Handles graceful degradation (warn by default)

2. TEST FILE CREATED (minimum 10 test cases)
   [ ] Created at .claude/hooks/<type>/<hook-name>.test.cjs
   [ ] Tests valid operations (3+ cases)
   [ ] Tests blocked operations (3+ cases)
   [ ] Tests edge cases (3+ cases)
   [ ] Tests error handling (1+ cases)

3. DOCUMENTATION UPDATED
   [ ] Added to .claude/hooks/README.md
   [ ] Documented trigger conditions
   [ ] Documented exit codes

4. REGISTRATION (BLOCKING)
   [ ] Added to settings.json (pre/post-tool hooks)
   [ ] Added to config.yaml (event hooks)
   [ ] Verified: grep "<hook-name>" .claude/settings.json

5. MEMORY UPDATED
   [ ] Added to learnings.md with hook summary

6. HOOK-AGENT MAP UPDATED (MANDATORY)
   [ ] Added new hook to @HOOK_AGENT_MAP.md Section 1 matrix
   [ ] Determined which agent archetypes are affected (based on hook trigger/tool target)
   [ ] Updated affected agents' `## Enforcement Hooks` sections
   [ ] Verified: `grep "<hook-name>" .claude/docs/@HOOK_AGENT_MAP.md || echo "ERROR: Hook not in agent map!"`

BLOCKING: If ANY item above is missing, hook creation is INCOMPLETE.

Step 8: Post-Creation Hook Registration (Phase 1 Integration)

This step is CRITICAL. After creating the hook artifact, you MUST register it in the hook discovery system.

Phase 1 Context: Phase 1 is responsible for tool and hook validation/discovery. Hooks created without registration are invisible to the system and will not be loaded at startup.

After hook file is written and tested:

1. Create/Update Hook Registry Entry in appropriate location: If registry doesn't exist, create .claude/context/artifacts/hook-registry.json: {"hooks": [{"name": "{hook-name}", "id": "{hook-name}", "description": "{Brief description from hook}", "category": "{safety|routing|memory|session|validation|audit}", "type": "{pre-tool|post-tool|user-prompt|session-start|session-end}", "version": "1.0.0", "targetTools": ["{Tool1}", "{Tool2}"], "enforcementMode": "{block|warn|off}", "defaultEnabled": true, "filePath": ".claude/hooks/{category}/{hook-name}.cjs", "testFilePath": ".claude/hooks/{category}/{hook-name}.test.cjs", "environmentVariable": "{HOOK_NAME}_MODE"}]}
2. Validate Hook Against Schema: Ensure hook validates against .claude/schemas/hook-schema.json (if exists): # Validate hook structure node -e " const hook = require('./.claude/hooks/{category}/{hook-name}.cjs'); if (hook.validate) console.log('✓ Has validate() export'); if (hook.PROJECT_ROOT) console.log('✓ Has PROJECT_ROOT'); if (hook.findProjectRoot) console.log('✓ Has findProjectRoot()'); "
3. Register Hook in Loader: Update .claude/lib/hooks/hook-loader.cjs (if exists) to include new hook: const HOOKS_MANIFEST = {'{hook-name}': {path: './.claude/hooks/{category}/{hook-name}.cjs', type: '{pre-tool|post-tool}', matcher: '{Bash|Write|Edit|Read}', // Optional enabled: true, enforcementMode: process.env.{HOOK_NAME}_MODE || 'warn'}};
4. Register Hook in Configuration: For pre/post-tool hooks - Update .claude/settings.json: {"hooks": {"pre-tool": ["./.claude/hooks/safety/{hook-name}.cjs"], "post-tool": ["./.claude/hooks/memory/{hook-name}.cjs"]}} For event hooks - Update .claude/config.yaml: hooks: UserPromptSubmit: - path:./.claude/hooks/routing/{hook-name}.cjs type: command SessionStart: - path:./.claude/hooks/session/{hook-name}.cjs type: command
5. Document in .claude/hooks/README.md: Add entry under appropriate category: ` #### {Hook Name} ({hook-name}.cjs) {Detailed description of what the hook does.} **When it runs:** {Trigger condition - e.g., "Before every Bash command", "After task completion"} **What it checks/does:** - {Check/action 1} - {Check/action 2} - {Check/action 3} **Enforcement mode:** process.env.{HOOK_NAME}_MODE (default: warn) **Test file:** .claude/hooks/{category}/{hook-name}.test.cjs **Related hooks:** {List any hooks that interact with this one} `
6. Update Memory: Append to .claude/context/memory/learnings.md: ## Hook: {hook-name} - **Type:** {pre-tool|post-tool|event} - **Category:** {safety|routing|memory|session|validation} - **Purpose:** {Detailed purpose} - **Trigger:** {When it runs} - **Enforcement:** {Block/warn/off by default} - **Integration Notes:** {Any special considerations}

Why this matters: Without hook registration:

• Hooks are not loaded at startup
• Hook validation doesn't occur
• System cannot discover available hooks
• Safety validators are bypassed
• "Invisible artifact" pattern emerges

Phase 1 Integration: Hook registry is the discovery mechanism for Phase 1, enabling the system to validate hooks against schema, load them at startup, and enforce safety rules consistently.

Step 9: Integration Verification (BLOCKING - DO NOT SKIP)

This step verifies the artifact is properly integrated into the ecosystem.

Before calling TaskUpdate({status: "completed"}), you MUST run the Post-Creation Validation workflow:

1. Run the 10-item integration checklist: node.claude/tools/cli/validate-integration.cjs.claude/hooks/<category>/<hook-name>.cjs
2. Verify exit code is 0 (all checks passed)
3. If exit code is 1 (one or more checks failed):

- Read the error output for specific failures - Fix each failure: - Missing hook registry -> Create registry entry (Step 8) - Missing settings.json entry -> Register hook (Step 8) - Missing documentation -> Add to hooks/README.md - Missing memory update -> Update learnings.md - Missing test file -> Create.test.cjs file - Re-run validation until exit code is 0

1. Only proceed when validation passes

This step is BLOCKING. Do NOT mark task complete until validation passes.

Why this matters: The Party Mode incident showed that fully-implemented artifacts can be invisible to the Router if integration steps are missed. This validation ensures no "invisible artifact" pattern.

Reference: .claude/workflows/core/post-creation-validation.md

CLI Reference

# Create hook using CLI tool
node .claude/tools/hook-creator/create-hook.mjs \
  --name "hook-name" \
  --type "PreToolUse|PostToolUse|UserPromptSubmit" \
  --purpose "Description of what the hook does" \
  --category "safety|routing|memory|audit|security|validation|custom" \
  --matcher "Edit|Write|Bash"  # Optional: tool matcher regex

# List all hooks
node .claude/tools/hook-creator/create-hook.mjs --list

# Validate hook structure
node .claude/tools/hook-creator/create-hook.mjs --validate "<path>"

# Assign to agents
node .claude/tools/hook-creator/create-hook.mjs --assign "name" --agents "agent1,agent2"

# Unregister hook
node .claude/tools/hook-creator/create-hook.mjs --unregister "<path>"

# Test with sample input
echo '{"tool_name":"Edit","tool_input":{"file_path":"test.js"} }' | node .claude/hooks/<category>/<hook-name>.cjs

Hook Patterns Reference

Pattern 1: Safety Validator (Pre-Tool)

For validating commands before execution:

'use strict';

/**
 * Validate Git Force Push
 * Prevents accidental force pushes to protected branches
 */

const PROTECTED_BRANCHES = ['main', 'master', 'production'];

function validate(context) {
  const { tool, parameters } = context;

  if (tool !== 'Bash') {
    return { valid: true, error: '' };
  }

  const command = parameters?.command || '';

  // Check for force push
  if (command.includes('git push') && (command.includes('--force') || command.includes('-f'))) {
    // Check if pushing to protected branch
    for (const branch of PROTECTED_BRANCHES) {
      if (command.includes(branch)) {
        return {
          valid: false,
          error: `Force push to ${branch} blocked. Use --force-with-lease instead.`,
        };
      }
    }

    return {
      valid: true,
      error: '',
      warning: 'Force push detected. Ensure you know what you are doing.',
    };
  }

  return { valid: true, error: '' };
}

module.exports = { validate };

Pattern 2: Memory Extractor (Post-Tool)

For extracting learnings after task completion:

'use strict';

/**
 * Extract Workflow Learnings
 * Automatically captures patterns from completed workflows
 */

const fs = require('fs');
const path = require('path');

function findProjectRoot() {
  let dir = __dirname;
  while (dir !== path.parse(dir).root) {
    if (fs.existsSync(path.join(dir, '.claude'))) return dir;
    dir = path.dirname(dir);
  }
  return process.cwd();
}

const LEARNINGS_PATH = path.join(findProjectRoot(), '.claude/context/memory/learnings.md');

function extractLearnings(context) {
  const { tool, parameters, result } = context;

  // Only process completed tasks
  if (!result || result.status !== 'completed') {
    return { extracted: false };
  }

  // Extract patterns from result
  const learnings = [];

  if (result.patterns) {
    learnings.push(...result.patterns);
  }

  if (result.decisions) {
    learnings.push(...result.decisions);
  }

  if (learnings.length === 0) {
    return { extracted: false };
  }

  // Append to learnings file
  const entry = `\n## [${new Date().toISOString().split('T')[0]}] ${context.taskName || 'Task'}\n\n`;
  const content = learnings.map(l => `- ${l}`).join('\n');

  fs.appendFileSync(LEARNINGS_PATH, entry + content + '\n');

  return { extracted: true, count: learnings.length };
}

module.exports = { extractLearnings };

Pattern 3: Routing Enforcer (User Prompt)

For enforcing routing protocols:

'use strict';

/**
 * Router First Enforcer
 * Ensures all requests go through the Router agent
 */

function validate(context) {
  const { prompt, currentAgent } = context;

  // Skip if already routed
  if (currentAgent === 'router') {
    return { valid: true, error: '' };
  }

  // Skip slash commands (handled by skill system)
  if (prompt && prompt.trim().startsWith('/')) {
    return { valid: true, error: '' };
  }

  // Suggest routing
  return {
    valid: true,
    error: '',
    warning: 'Consider using Router to spawn appropriate agent via Task tool.',
  };
}

module.exports = { validate };

Pattern 4: Session Initializer

For session lifecycle management:

'use strict';

/**
 * Session Memory Initializer
 * Reminds agents to read memory files at session start
 */

const fs = require('fs');
const path = require('path');

function findProjectRoot() {
  let dir = __dirname;
  while (dir !== path.parse(dir).root) {
    if (fs.existsSync(path.join(dir, '.claude'))) return dir;
    dir = path.dirname(dir);
  }
  return process.cwd();
}

const MEMORY_FILES = [
  '.claude/context/memory/learnings.md',
  '.claude/context/memory/issues.md',
  '.claude/context/memory/decisions.md',
];

function initialize() {
  const root = findProjectRoot();

  console.log('\n' + '='.repeat(50));
  console.log(' SESSION MEMORY REMINDER');
  console.log('='.repeat(50));
  console.log('\n  Before starting work, read these memory files:');

  for (const file of MEMORY_FILES) {
    const fullPath = path.join(root, file);
    if (fs.existsSync(fullPath)) {
      const stats = fs.statSync(fullPath);
      const modified = stats.mtime.toISOString().split('T')[0];
      console.log(`  - ${file} (updated: ${modified})`);
    }
  }

  console.log('\n' + '='.repeat(50) + '\n');

  return { initialized: true };
}

// Run on direct execution
if (require.main === module) {
  initialize();
}

module.exports = { initialize };

Examples

Security Validation Hook

node .claude/tools/hook-creator/create-hook.mjs \
  --name "secret-detector" \
  --type "PreToolUse" \
  --purpose "Blocks commits containing secrets or credentials" \
  --category "security" \
  --matcher "Bash"

Audit Logging Hook

node .claude/tools/hook-creator/create-hook.mjs \
  --name "operation-logger" \
  --type "PostToolUse" \
  --purpose "Logs all file modifications to audit trail" \
  --category "audit"

Intent Analysis Hook

node .claude/tools/hook-creator/create-hook.mjs \
  --name "intent-classifier" \
  --type "UserPromptSubmit" \
  --purpose "Classifies user intent for intelligent routing" \
  --category "routing"

Workflow Integration

This skill is part of the unified artifact lifecycle. For complete multi-agent orchestration:

Router Decision: .claude/workflows/core/router-decision.md

• How the Router discovers and invokes this skill's artifacts

Artifact Lifecycle: .claude/workflows/core/skill-lifecycle.md

• Discovery, creation, update, deprecation phases
• Version management and registry updates
• CLAUDE.md integration requirements

External Integration: .claude/workflows/core/external-integration.md

• Safe integration of external artifacts
• Security review and validation phases

Cross-Reference: Creator Ecosystem

This skill is part of the Creator Ecosystem. After creating a hook, consider if companion artifacts are needed:

Integration Workflow

After creating a hook that needs additional capabilities:

// 1. Hook created but needs dedicated skill
Skill({ skill: 'skill-creator' });
// Create skill that encapsulates hook logic

// 2. Hook needs to be assigned to agent
// Update agent's workflow to include hook awareness

// 3. Hook needs workflow for testing
// Create workflow in .claude/workflows/<hook-name>-test-workflow.md

Post-Creation Checklist for Ecosystem Integration

After hook is fully created and validated:

[ ] Does hook need a skill wrapper? -> Use skill-creator
[ ] Does hook need dedicated agent? -> Use agent-creator
[ ] Does hook need testing workflow? -> Create workflow
[ ] Should hook be enabled by default? -> Update config.yaml
[ ] Does hook interact with other hooks? -> Document in README.md
[ ] Run post-creation validation -> node .claude/tools/cli/validate-integration.cjs .claude/hooks/<category>/<hook-name>.cjs

Iron Laws of Hook Creation

These rules are INVIOLABLE. Breaking them causes silent failures.

1. NO HOOK WITHOUT validate() EXPORT
   - Every hook MUST export validate() function
   - Hooks without validate() cannot be called programmatically

2. NO HOOK WITHOUT main() FOR CLI
   - Every hook MUST have main() for CLI execution
   - Run only when require.main === module

3. NO HOOK WITHOUT ENFORCEMENT CONTROLS
   - Support 'block|warn|off' via environment variable
   - Default to 'warn' unless explicitly required to block
   - Never crash on malformed input

4. NO HOOK WITHOUT ERROR HANDLING
   - Wrap JSON.parse in try/catch
   - Handle missing parameters gracefully
   - Return valid: true when unsure (fail open, not closed)

5. NO HOOK WITHOUT TEST FILE
   - Every hook needs <hook-name>.test.cjs
   - Test valid, invalid, and edge cases

6. NO HOOK WITHOUT DOCUMENTATION
   - Add to .claude/hooks/README.md
   - Document trigger, behavior, exit codes

7. CROSS-PLATFORM PATHS
   - Use path.join() not string concatenation
   - Handle both / and \ path separators
   - Use path.normalize() for comparison

8. NO CREATION WITHOUT SYSTEM IMPACT ANALYSIS
   - Check if hook requires settings.json registration
   - Check if hook requires config.yaml registration
   - Update @HOOK_AGENT_MAP.md with new hook row (MANDATORY)
   - Update affected agents' Enforcement Hooks sections (MANDATORY)
   - Check if related hooks need updating
   - Document all system changes made

9. IRON LAW I: PRE-TOOL HOOKS MUST VALIDATE AGAINST JSON SCHEMA
   - PreToolUse hooks MUST compile and validate input against the companion
     schemas/input.schema.json using AJV or equivalent before allowing execution
   - Pattern: compile(schema) → validate(input) → exit 2 on schema failure
   - Never block (exit 2) on schema-load errors — fail open, not closed
   - Search: site:github.com "preToolUse" "ajv" "validate" filetype:cjs

10. IRON LAW III: POST-TOOL HOOKS MUST EMIT OBSERVABILITY EVENTS
    - PostToolUse hooks MUST append a structured JSON line to
      .claude/context/runtime/tool-events.jsonl via the centralized emitter:
      const { sendEvent } = require('.claude/tools/observability/send-event.cjs')
    - Required fields: tool_name, agent_id, session_id, outcome, timestamp
    - Never crash on emit failure (try/catch, fail open)
    - Inspect events: node .claude/tools/observability/send-event.cjs --tail 20

Integration Points

• Ecosystem Assessor: Hook creator integrates with ecosystem assessment for reverse lookups
• Agent Creator: Agents can reference hooks in their frontmatter
• Skill Creator: Skills can define hooks in their hooks/ directory
• Settings.json: Hooks are auto-registered with proper triggers and matchers
• Config.yaml: Event hooks registered for UserPromptSubmit, SessionStart, etc.

Architecture Compliance

File Placement (ADR-076)

• Hooks: .claude/hooks/{category}/ (safety, routing, memory, session, validation, audit)
• Hook tests: .claude/hooks/{category}/{name}.test.cjs (co-located with hooks)
• Tests: tests/ (integration tests for hooks)
• Related docs: .claude/docs/
• Hook registry: .claude/hooks/README.md

Documentation References (CLAUDE.md v2.2.1)

• Reference files use @notation: @ENFORCEMENT_HOOKS.md, @TOOL_REFERENCE.md
• Located in: .claude/docs/@*.md
• See: CLAUDE.md Section 1.3 (ENFORCEMENT HOOKS reference)

Shell Security (ADR-077)

• NEW SAFETY HOOKS: bash-cwd-validator.cjs, shell-injection-validator.cjs, variable-quoting-validator.cjs (ADR-077 Phase 2)
• PHASE 3 HOOKS: shellcheck-validator.cjs, command-allowlist-validator.cjs (reference implementations)
• Hook tests MUST validate shell security patterns
• See:.claude/docs/SHELL-SECURITY-GUIDE.md
• Apply to: all safety hooks, pre-tool hooks, validation hooks

Recent ADRs

• ADR-075: Router Config-Aware Model Selection
• ADR-076: File Placement Architecture Redesign
• ADR-077: Shell Command Security Architecture

File Placement & Standards

Output Location Rules

This skill outputs to: .claude/hooks/<category>/

Categories:

• safety/ - Safety validators (command validation, security checks, shell security)
• routing/ - Router enforcement hooks
• memory/ - Memory management hooks
• session/ - Session lifecycle hooks
• validation/ - Input/output validation hooks

Mandatory References

• File Placement: See .claude/docs/FILE_PLACEMENT_RULES.md
• Developer Workflow: See .claude/docs/DEVELOPER_WORKFLOW.md
• Artifact Naming: See .claude/docs/ARTIFACT_NAMING.md

Enforcement

File placement is enforced by file-placement-guard.cjs hook. Invalid placements will be blocked in production mode.

Memory Protocol (MANDATORY)

Before creating a hook:

cat .claude/context/memory/learnings.md

Check for:

• Previously created hooks
• Known hook patterns
• User preferences for hook behavior

After completing:

• New hook created -> Append to .claude/context/memory/learnings.md
• Issue with hook -> Append to .claude/context/memory/issues.md
• Hook design decision -> Append to .claude/context/memory/decisions.md

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.

Ecosystem Alignment Contract (MANDATORY)

This creator skill is part of a coordinated creator ecosystem. Any artifact created here must align with and validate against related creators:

• agent-creator for ownership and execution paths
• skill-creator for capability packaging and assignment
• tool-creator for executable automation surfaces
• hook-creator for enforcement and guardrails
• rule-creator and semgrep-rule-creator for policy and static checks
• template-creator for standardized scaffolds
• workflow-creator for orchestration and phase gating
• command-creator for user/operator command UX

Cross-Creator Handshake (Required)

Before completion, verify all relevant handshakes:

1. Artifact route exists in .claude/CLAUDE.md and related routing docs.
2. Discovery/registry entries are updated (catalog/index/registry as applicable).
3. Companion artifacts are created or explicitly waived with reason.
4. validate-integration.cjs passes for the created artifact.
5. Skill index is regenerated when skill metadata changes.

Research Gate (Exa + arXiv — BOTH MANDATORY)

For new patterns, templates, or workflows, research is mandatory:

1. Use Exa for implementation and ecosystem patterns:

- mcp__Exa__web_search_exa({query: '<topic> 2025 best practices'}) - mcp__Exa__get_code_context_exa({query: '<topic> implementation examples'})

1. Search arXiv for academic research (mandatory for AI/ML, agents, evaluation, orchestration, memory/RAG, security):

- Via Exa: mcp__Exa__web_search_exa({query: 'site:arxiv.org <topic> 2024 2025'}) - Direct API: WebFetch({url: 'https://arxiv.org/search/?query=<topic>&searchtype=all&start=0'})

1. Record decisions, constraints, and non-goals in artifact references/docs.
2. Keep updates minimal and avoid overengineering.

arXiv is mandatory (not fallback) when topic involves: AI agents, LLM evaluation, orchestration, memory/RAG, security, static analysis, or any emerging methodology.

Regression-Safe Delivery

• Follow strict RED -> GREEN -> REFACTOR for behavior changes.
• Run targeted tests for changed modules.
• Run lint/format on changed files.
• Keep commits scoped by concern (logic/docs/generated artifacts).

Optional: Evaluation Quality Gate

Run the shared evaluation framework to verify hook quality:

node .claude/skills/skill-creator/scripts/eval-runner.cjs --skill hook-creator

Grader assertions for hook artifacts:

• Exit codes correct: Hook exits 0 (allow) or 2 (block) only; exit 1 is never used (treated as error, not block per SE-03)
• 100ms performance budget: Hook body completes in under 100ms; no network calls, no blocking I/O, no long computation in the hot path
• Fail-open vs fail-closed policy: Security hooks (routing, creator, write) are fail-closed (process.exit(2) on errors); advisory and PostToolUse hooks are fail-open (process.exit(0) on errors)
• Graceful error handling: Hook body is wrapped in try/catch; unexpected errors exit 0 (non-critical) or 2 (security-critical) — never crash without an exit code
• Registration complete: Hook is registered in .claude/settings.json and documented in @ENFORCEMENT_HOOKS.md

See .claude/skills/skill-creator/EVAL_WORKFLOW.md for full evaluation protocol and grader/analyzer agent usage.