writing-skills

To bulletproof agent skill documentation against pressure scenarios, apply a red-green-refactor workflow that captures baseline failing scenarios and adds explicit counters for agent rationalizations. This approach catches documentation loopholes before deployment.

Rationalization-proofing in skill authoring captures baseline agent excuses verbatim and adds explicit counters and red flags to documentation. This prevents AI agents from skipping required process discipline when facing time, authority, sunk-cost, or exhaustion pressures.

Optimizing SKILL.md metadata for reliable agent discovery requires third-person YAML frontmatter with a name and description, plus CSO-aligned "when to use" descriptions. This ensures the right loading behavior is triggered when agents search for relevant skills.

Test-driven development for skill documentation replaces "write-it-and-hope" with pressure-scenario validation. You capture a baseline failing scenario, write minimal corrective instructions, and iteratively close loopholes with re-verification to ensure compliance.

The best way to prevent AI agents from skipping process discipline is forcing a red-green-refactor approach with rationalization-proofing during skill authoring. This captures baseline agent excuses and adds explicit counters before the skill is deployed.