rulehook — Three-Layer Compliance Audit System

⚠️ READ FIRST — MANDATORY

Before installing, understand: rulehook enforces rules by watching your Claude agent's behavior. It will remind you when you violate your own rules.

After installation: copy hooks/ scripts to ~/.claude/scripts/hooks/, register in ~/.claude/settings.json, and add the L1 prevention block to any document-generation skill.

Quick Install

# 1. Copy hooks
cp hooks/stop-doc-template-check.js ~/.claude/scripts/hooks/
cp hooks/session-start-compliance-audit.js ~/.claude/scripts/hooks/

# 2. Merge config/settings-snippet.json into ~/.claude/settings.json
# 3. Add PRE-GENERATION-CHECK to relevant skills (see skills/PRE-GENERATION-CHECK.md)
# 4. Verify
node tests/simulate-all.js

What Problem This Solves

Claude Code has 20+ rules in ~/.claude/rules/. Rules are soft constraints — the agent reads them but skips them under engineering momentum. The result: you keep reminding Claude "I told you last time."

rulehook bridges hard hooks with soft rules through three defence lines:

1. L1 Prevention — SKILL.md injection: "Before generating, ask the user which template"
2. L2 Intercept — Stop hook two-stage grep: scan transcript → warn on violation
3. L3 Warning — SessionStart compliance audit: "You scored 40% on this rule last week"

Zero token cost. Zero external dependencies. Node.js stdlib only.

Architecture

User says "make a Word doc"
  → L1 fires (skill PRE-GENERATION CHECK)
  → Agent generates docx
  → Stop hook: grep transcript → violation? → systemMessage
  → Next session: SessionStart reads observations.jsonl → "40% compliance — be careful"

Adding Rules

Edit hooks/session-start-compliance-audit.js → RULES object. One JSON entry per rule.

Documentation

Full design doc: docs/合规审计系统-三层防线机制.md
Test suite: tests/