tractatus/docs/ENFORCEMENT_AUDIT.md

Governance Enforcement Audit

Problem Statement

Claude missed "ffs" trigger (inst_082) and token checkpoint monitoring (inst_075) despite both being active HIGH persistence instructions. This reveals a voluntary compliance failure pattern.

Root Cause

Instructions marked MANDATORY but lacking architectural enforcement mechanisms rely on voluntary compliance, which fails under cognitive load or context pressure.

Enforcement Analysis: 54 HIGH Persistence Instructions

✅ ALREADY ENFORCED (8 instructions)

These have architectural hooks/scripts that BLOCK or ALERT automatically:

1. inst_038: Pre-action checks → .claude/hooks/framework-audit-hook.js
2. inst_065: Session initialization → scripts/session-init.js (mandatory)
3. inst_070: Secret detection → Git pre-commit hook (if installed)
4. inst_071: Pre-deployment checklist → Partially automated
5. inst_075: Token checkpoints → .claude/hooks/check-token-checkpoint.js ✨ NEW
6. inst_077: Session closedown → scripts/session-closedown.js
7. inst_078: "ff" trigger → .claude/hooks/trigger-word-checker.js ✨ NEW
8. inst_082: "ffs" trigger → .claude/hooks/trigger-word-checker.js ✨ NEW

⚠️ NEEDS ENFORCEMENT (12 critical instructions)

HIGH PRIORITY - Implement Now

inst_027: NEVER modify instruction-history.json without approval

• Risk: HIGH - Could corrupt governance system
• Enforcement: PreToolUse hook to block Write/Edit on .claude/instruction-history.json
• Implementation: Add to existing framework-audit-hook.js

inst_064: Framework components MUST be actively used

• Risk: HIGH - Framework fade = governance collapse
• Enforcement: session-state.json tracking + periodic verification
• Implementation: Add activity report to session-init output

inst_066: Conventional commit format required

• Risk: MEDIUM - Makes git history hard to parse
• Enforcement: Git commit-msg hook to validate format
• Implementation: Create .git/hooks/commit-msg

inst_068: Test requirements before commits/deployments

• Risk: HIGH - Broken code reaches production
• Enforcement: Pre-commit hook to run relevant tests
• Implementation: Add to git pre-commit hook

inst_023: Background process management required

• Risk: MEDIUM - Orphaned processes waste resources
• Enforcement: Track spawned processes in session-state.json
• Implementation: Hook on Bash tool with run_in_background=true

inst_046: Security events MUST be logged

• Risk: HIGH - Undetected attacks
• Enforcement: Audit log verification in pre-deployment checklist
• Implementation: Add automated check to deployment script

MEDIUM PRIORITY - Implement Later

inst_040: "all" command verification (needs workflow enforcement) inst_056: Batch operation validation (needs workflow tracking) inst_019: ContextPressureMonitor must account for total context (already doing this)

📋 INFORMATIONAL/GUIDELINES (34 instructions)

These provide context, standards, or principles that don't require automated enforcement:

• SYSTEM quadrant (9): Database ports, CSP rules, security standards, credential handling
• STRATEGIC quadrant (17): Quality standards, approval requirements, values principles, architectural guidance
• OPERATIONAL quadrant (8): Best practices, deployment procedures, file management

These are enforced through:

• Code review
• Human approval gates (inst_005)
• Documentation and training
• Framework service checks (BoundaryEnforcer, CrossReferenceValidator)

Enforcement Tiers

Tier 1: Architectural Blocks (MANDATORY)

Cannot proceed without compliance. Examples:

• session-init.js blocks without local server
• Pre-action hooks block dangerous operations
• Secret detection prevents commits with credentials

Tier 2: Architectural Alerts (CRITICAL)

Warns immediately but allows override with explicit approval. Examples:

• Token checkpoint warnings
• Trigger word reminders
• CSP compliance notifications

Tier 3: Audit Trails (MONITORING)

Logs violations for review but doesn't block. Examples:

• Framework component usage tracking
• Security event logging
• Background process inventory

Tier 4: Human Gates (APPROVAL)

Requires explicit human decision. Examples:

• Architectural changes (inst_005)
• Values-sensitive content (BoundaryEnforcer)
• Production deployments (inst_071)

Implementation Priority Queue

Immediate (This Session)

1. ✅ Token checkpoint monitoring hook
2. ✅ Trigger word detection hook
3. ⏳ instruction-history.json write protection
4. ⏳ Framework activity verification
5. ⏳ Conventional commit format hook

Next Session

6. Test requirement enforcement (pre-commit)
7. Background process tracking
8. Security event audit verification
9. Batch operation workflow tracking

Ongoing

10. Continuous monitoring of framework fade
11. Periodic enforcement effectiveness review
12. New instruction classification and enforcement design

Meta-Enforcement System

Create scripts/audit-enforcement.js that:

1. Scans all HIGH persistence instructions
2. Identifies those with MUST/NEVER/MANDATORY language
3. Checks if corresponding enforcement mechanism exists
4. Alerts when new imperative instructions lack enforcement
5. Generates enforcement gap report

Run automatically:

• At session start (via session-init.js)
• When instruction-history.json changes
• On demand via "ffs" or dedicated trigger

Philosophy

"If it's MANDATORY, it must be ENFORCED architecturally, not documented."

Voluntary compliance works under ideal conditions but fails under:

• Cognitive load (complex multi-step tasks)
• Context pressure (approaching token limits)
• Auto-compact recovery (instruction injection timing)
• Session handoffs (knowledge discontinuity)

Architectural enforcement (hooks, scripts, blocks) is reliable regardless of conditions.

---

Created: 2025-10-25 Trigger: Missed "ffs" code word recognition Outcome: Systematic enforcement architecture design