# Trigger Word Enforcement Architecture

## Problem
Claude failed to recognize "ffs" code word despite inst_082 being active. This was a **trigger recognition failure**, not a missing instruction issue.

## Root Cause
- inst_082 (ffs trigger) exists and is marked active: true
- inst_078 (ff trigger) exists and is marked active: true
- Both are HIGH persistence, SYSTEM quadrant
- scripts/framework-stats.js and scripts/framework-audit-response.js both exist
- **BUT**: No architectural enforcement to ensure Claude checks for trigger words on EVERY user message

## Solution
Created `.claude/hooks/trigger-word-checker.js` that runs on every user prompt submission:

```javascript
// Detects:
// - "ffs" (exact match or word boundary) → framework-stats.js
// - "ff " (prefix) → framework-audit-response.js
```

Registered in `.claude/settings.json`:
```json
"UserPromptSubmit": [
  {
    "hooks": [
      {
        "type": "command",
        "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/trigger-word-checker.js",
        "timeout": 2
      }
    ]
  }
]
```

## How It Works

1. **User types "ffs"**
2. Hook runs BEFORE Claude processes the message
3. Hook outputs: `⚠️ CODE WORD DETECTED: "ffs"`
4. Hook provides instruction: `Claude MUST run: node scripts/framework-stats.js`
5. Claude sees the hook output and executes the command

## Testing

```bash
# Test ffs trigger
node .claude/hooks/trigger-word-checker.js '{"text":"ffs"}'
# Output: ⚠️ CODE WORD DETECTED: "ffs"
#         Claude MUST run: node scripts/framework-stats.js

# Test ff trigger
node .claude/hooks/trigger-word-checker.js '{"text":"ff what are the boundary rules?"}'
# Output: ⚠️ CODE WORD DETECTED: "ff"
#         Claude MUST run: node scripts/framework-audit-response.js --prompt "..." --type "boundary_question"

# Test normal message (no output)
node .claude/hooks/trigger-word-checker.js '{"text":"please fix the bug"}'
# (silent - no trigger detected)
```

## Philosophy
**Governance is enforced architecturally, not documented.**

Relying on voluntary compliance (Claude remembering to check for triggers) = framework fade.
Architectural enforcement (hook that runs on EVERY message) = reliable governance.

## Files Modified
- `.claude/hooks/trigger-word-checker.js` - New hook script
- `.claude/settings.json` - Registered UserPromptSubmit hook

## References
- inst_082: ffs trigger definition
- inst_078: ff trigger definition
- CLAUDE.md lines 48-64: ff documentation
- CLAUDE.md lines 66-88: ffs documentation

## Future Enhancements
If Claude Code doesn't support UserPromptSubmit hooks natively:
1. Add to session-init.js output as prominent reminder
2. Create a PreToolUse hook that checks user's last message
3. Add to system prompt injection if hook system insufficient

---
**Created**: 2025-10-25
**Issue**: Claude didn't recognize "ffs" code word
**Fix**: Architectural hook enforcement