Agent Skill Pre-Flight Checklist

Most agent skills are sold on their instructions. Instructions are not behavior — but before you pay for a full blind behavior test, there is a fast static check that catches the defects most likely to sink a skill.

The Pre-Flight Checklist runs 12 pass/fail checks on any SKILL.md:

• Is the behavior specified in testable terms, or just described?
• Are decision tiers and thresholds defined — and are their boundaries unambiguous?
• Is the decision vocabulary consistent throughout?
• Are the safety rules explicit: no fabricated verification, no leaking secrets from inputs, no obeying instructions embedded in processed material?
• Are scope limits and failure conditions stated honestly?
• Is the output format disciplined enough to grade?

You get a score, an evidence-quoted verdict per check, and the top fixes in impact order.

Honest limits: this is a static text review. It reads the skill; it does not execute it. A skill can pass all 12 checks and still misbehave in a real run — that is exactly what happened in the documented case behind this method (a well-built skill, clean on structure, scored 7/12 on blind behavioral tests until one calibration rule was fixed, then 12/12).

When you need executed proof, the full method is the Agent Skill Behavior Test Lab ($14.99, also on Agensi): blind case generation, frozen expectations, real runner execution, deterministic grading, safety sub-scores, and a release-gate verdict.

If this checklist helped you catch something before you shipped, an honest review helps other builders find it — good, bad, or mixed.