writing-skills

Agent skill authoring requires concise metadata, strong trigger wording, and iterative testing to close rationalization loopholes. Structuring SKILL.md with compact instructions and explicit counters ensures agents comply reliably under pressure instead of ignoring or inconsistently following instructions.

Testing agent skills before deployment involves running baseline scenarios and pressure tests to expose failure modes. Refactoring cycles validate behavior under time pressure, sunk cost, and authority pressure, ensuring the skill documentation holds up reliably before making it available.

Agents ignore skill documentation when metadata lacks clear triggers or concise instructions. Rationalization loopholes allow agents to bypass rules under pressure; adding explicit counters, red flags, and discipline-enforcing wording hardens the skill and forces consistent compliance.

Hardening discipline-enforcing skills requires adding explicit counters, red flags, and rationalization handling directly within the SKILL.md. By anticipating pressure scenarios like sunk cost or authority pressure, you close loopholes that allow agents to bypass the intended rules.

Organizing heavy reference material in agent skills requires splitting complex content into supporting files. Keeping the main skill focused and token-efficient ensures agents process the core instructions without being overwhelmed by excessive context.