majiayu000 User Guide: mastering_hooks_cch
The Intelligent Policy Author for Code Agent Context Hooks
The mastering_hooks_cch skill is an intelligent assistant that helps you design, generate, and manage AI policies for Claude using the Claude Context Hooks (CCH) system.
Mental model: • CCH CLI (cch) → Policy engine (deterministic runtime) • mastering_hooks_cch → Policy author (intelligent designer)
The skill does not run at runtime. It runs on-demand to help you create, modify, and explain policies.
⸻
Table of Contents 1. What the Skill Does 2. Getting Started 3. Core Capabilities 4. Common Workflows 5. Governance & Policy Modes 6. Provenance & Explainability 7. Troubleshooting & Analysis 8. Trigger Cheat Sheet 9. Mental Model (Important)
⸻
What the Skill Does
The mastering_hooks_cch skill is responsible for: • Installing and verifying the CCH binary • Analyzing your project structure • Discovering existing skills • Translating human rules into enforceable policies • Generating hooks.yaml, validators, and context templates • Explaining why policies exist and how they behave • Debugging policy failures using logs
It acts as a policy consultant + architect + debugger.
It never executes hooks itself.
⸻
Getting Started
You never need to write YAML manually unless you want to.
- Install CCH
Just ask Claude:
“Install CCH”
The skill will: • Detect OS and architecture • Download the correct binary • Verify checksums • Install it to: • project (.claude/bin/cch) or • global user path
It also verifies compatibility using:
cch --version --json
⸻
- Initial Project Setup
To scaffold policies for a new repository:
“Set up hooks for this project”
The skill will: 1. Scan your project 2. Detect: • languages • frameworks • infra tools • CI/CD workflows 3. Discover skills in .claude/skills/ 4. Parse CLAUDE.md 5. Generate: • .claude/hooks.yaml • .claude/context/.md • .claude/validators/.py
Then it asks for your approval before installing.
⸻
Core Capabilities
Project Analysis
The skill builds a real understanding of your system:
Skill Discovery Scans:
.claude/skills/**/SKILL.md
And extracts: • file types • directories • tools used
Used to auto-wire skill triggers.
⸻
Rule Extraction Reads CLAUDE.md and converts:
Phrase Interpretation MUST / NEVER Enforce (block) SHOULD / PREFER Warn AVOID Warn or Audit
Into real enforcement logic.
This is the core value:
turning human intent into deterministic policy.
⸻
Safety & Risk Detection The skill detects: • .env files • secrets directories • terraform/, cdk/ • .github/workflows/
And suggests: • production guards • destructive command blocks • credential exposure policies
⸻
Conflict Detection
If your policies contradict each other, the skill detects it.
Example:
“Use functional style” “Use OOP everywhere”
The skill will: • flag conflict • explain the contradiction • ask you to choose precedence
This prevents policy paradoxes.
⸻
Common Workflows
- Enforcing Coding Standards
User:
“Help me enforce the no-console-log rule”
Skill: • Parses CLAUDE.md • Creates validator script • Adds rule to hooks.yaml • Asks: block or warn?
Result:
Your rule becomes executable policy, not a suggestion.
⸻
- Context Injection (Skill Wiring)
User:
“Trigger the CDK skill when editing infra files”
Skill: • Locates aws-cdk/SKILL.md • Detects relevant paths • Creates PreToolUse rule • Injects CDK docs automatically
Now Claude always has the right playbook.
⸻
- Safety Guardrails
User:
“Block force pushes”
Skill: Creates:
- name: block-force-push tools: [Bash] command_match: "git push.*(--force|-f)" mode: enforce
This is real-time protection.
⸻
- Interactive Policy Design
User:
“Help me configure hooks”
The skill runs a guided interview: • strict vs relaxed? • security-sensitive? • infra-heavy? • solo or team?
Then generates a custom policy profile.
⸻
Governance & Policy Modes
CCH supports real policy governance.
When creating rules, you can specify a mode:
Policy Modes
Mode Behavior Enforce Block if violated Warn Inject warning Audit Log only
Example:
“Warn me if I edit production files”
This lets you: • dry-run new policies • roll out gradually • observe before enforcing
This is how real organizations deploy policy.
⸻
Provenance & Explainability
Every rule created by the skill includes metadata:
metadata: author: user created_by: mastering_hooks_cch reason: "From CLAUDE.md" confidence: high last_reviewed: 2025-01-21
You can ask:
“Why does this rule exist?” “Who created this rule?” “How confident is this policy?”
The skill will show: • source • rationale • history • trigger frequency
This turns CCH into a self-documenting policy system.
⸻
Troubleshooting & Analysis
Log Analysis
Runtime logs live at:
~/.claude/logs/cch.log
User:
“Why isn’t my hook working?”
Skill: • reads logs • finds matching event • explains: • matcher failed? • wrong extension? • audit mode? • overridden by higher priority rule?
This is root cause analysis for AI behavior.
⸻
Explain a Rule
User:
“Explain the no-console-log rule”
Skill shows: • matchers • actions • mode • priority • provenance • last triggers
This mirrors:
cch explain rule no-console-log
But in natural language.
⸻
Trigger Cheat Sheet
These phrases activate the skill:
Intent Example Install “Install CCH” Setup “Set up hooks for this project” Add rule “Create a hook for…” Enforce “Enforce rule with hook” Wire skill “Trigger skill with hook” Debug “Why isn’t my hook working?” Explain “Explain this rule” Validate “Validate hooks” Remove “Remove hook” Update “Update hooks”
The skill may also proactively suggest policies when: • you add new skills • you edit CLAUDE.md • you attempt risky actions
⸻
Mental Model (Important)
What runs when?
Component When mastering_hooks_cch On demand CCH CLI (cch) On every hook event hooks.yaml Loaded every event validators Executed conditionally logs Written always
⸻
The Core Philosophy
LLMs do not enforce policy. LLMs are subject to policy.
The skill: • designs policy
The CLI: • enforces policy
Claude: • operates inside policy
This makes CCH fundamentally different from: • prompt engineering • system messages • “please follow rules” instructions
CCH is real governance for AI systems.
⸻
One-Line Summary
The mastering_hooks_cch skill is not a chatbot.
It is:
An intelligent policy authoring system for building deterministic, auditable, and enforceable AI behavior.
It lets you treat Claude not as a magical assistant…
…but as a governed system with real rules.