writing-skills

Test skills documentation by running a RED-GREEN-REFACTOR loop to baseline failure cases, write minimal guidance, and close loopholes before deployment. This enforces a mandatory stop-and-test checklist to ensure bulletproof docs.

The RED-GREEN-REFACTOR workflow for skill writing is a test-driven loop: run a failing baseline to capture failure modes, write a minimal GREEN skill to pass it, and complete a REFACTOR pass to tighten loopholes and ensure deployment readiness.

Manage agent rationalizations by capturing them in a formal rationalization table and flagging them with red flags. This structured approach addresses agent excuses and prevents regressions by documenting explicit countermeasures.

Prevent regressions in Skills documentation by baselining failure cases through test-driven writing. This approach reduces ambiguity and documents explicit countermeasures to ensure bulletproof guidance across the repository.

Test-driven skill writing supports references to best practice background materials like anthropic-best-practices.md and testing-skills-with-subagents.md to align documentation with established guidelines and meta-testing evaluation patterns.

Avoid test-driven skill writing when you need rapid, informal prototyping without deployment hygiene. This approach enforces explicit rationalizations, red-flag tables, and mandatory stop-and-test checklists, which may overhead simple documentation tasks.