Bulletproof Skill

Bulletproof Skill Command

use_when:

• Hardening skills against rationalization and bypass behaviors
• Identifying loopholes in skill language
• Generating rationalization tables
• Creating red flags lists
• Preparing skills for production

do_not_use_when:

• Testing skill functionality - use /test-skill instead
• Evaluating skill quality - use /skills-eval instead
• Creating new skills - use /create-skill instead

Systematically hardens skills against rationalization and bypass behaviors.

Usage

# Analyze skill for loopholes and rationalizations
/bulletproof-skill skills/my-skill

# Generate full bulletproofing report
/bulletproof-skill skills/my-skill --report

# Apply suggested fixes automatically (interactive)
/bulletproof-skill skills/my-skill --interactive

# Check specific aspects only
/bulletproof-skill skills/my-skill --check loopholes
/bulletproof-skill skills/my-skill --check rationalizations
/bulletproof-skill skills/my-skill --check red-flags

What is Bulletproofing?

Problem: LLMs rationalize around constraints through:

• Vague language interpretation ("usually", "try to", "generally")
• "Spirit vs letter" compliance
• Perceived simplicity exemptions
• Memory-based shortcuts

Solution: Makes implicit requirements explicit, adds rationalization counters, closes loopholes.

Workflow

Step 1: Loophole Analysis

Scans skill content for weakness patterns:

• Vague Language: "usually", "try to", "generally", "should"
• Missing Exceptions: Undefined terms like "complex", "important"
• Ambiguous Conditions: Subjective criteria allowing bypass
• Escape Hatches: "Skip if not needed" without criteria

Step 2: Rationalization Detection

Identifies common bypass patterns:

• Simplicity Bypass: "This is just a simple task"
• Memory Shortcut: "I remember what the skill says"
• Overkill Excuse: "The skill is overkill for this"
• Deferral: "Let me do this one thing first"
• Spirit vs Letter: "Technically following the approach"

Step 3: Generate Rationalization Table

Creates comprehensive mapping of rationalizations → counters.

Step 4: Create Red Flags List

Builds self-check questions to catch bypass attempts.

Step 5: Suggest Counters

Provides explicit anti-rationalization language.

Severity Levels

• Critical: Allows complete skill bypass (MUST fix)
• High: Enables significant deviation (SHOULD fix)
• Medium: Potential for selective application (CONSIDER fixing)
• Low: Minor wording improvements (OPTIONAL)

Output

Summary Report

LOOPHOLE ANALYSIS
Skill: skills/my-skill v0.2.0

VAGUE LANGUAGE (4 instances)
- Line 23: "Usually follow..." → "ALWAYS follow..."
- Line 47: "Try to establish..." → "MUST establish..."

Total Issues: 10 (Critical: 2, High: 5, Medium: 1, Low: 1)
Recommendation: Address all CRITICAL and HIGH before production

Rationalization Table

| Thought | Reality | Counter |
|---------|---------|---------|
| "This is just simple" | Skills prevent scope creep | Apply EVERY time |
| "I remember the skill" | Memory drifts | Read CURRENT version |

Red Flags List

## Red Flags - STOP and Check Skill
- "This is just a simple..."
- "I remember what it says..."
- "This doesn't need the full..."

Detailed Guides

For comprehensive patterns and examples, see the bulletproof-skill source code and the abstract methodology framework.

Best Practices

When to Bulletproof:

• Before skill goes to production
• After discovering bypass instances
• When skill compliance is critical
• During skill refactoring

What Makes Good Bulletproofing:

• Specific, measurable counters
• Complete rationalization coverage
• Clear red flags
• Actionable fixes

Common Mistakes:

• Over-bulletproofing simple skills
• Missing common rationalizations
• Vague counters that invite new rationalizations
• Not testing bulletproofed version

Integration

# Typical workflow
/create-skill my-skill          # Create skill
/test-skill my-skill            # Test functionality
/bulletproof-skill my-skill     # Harden against bypass
/skills-eval my-skill           # Final validation

See Also

• /create-skill: Create new skills
• /test-skill: Test skill functionality
• /skills-eval: Evaluate skill quality
• abstract:methodology-curator: Browse anti-rationalization frameworks