Skill Activation Test
QA gate for your own Agent Skills — not for third-party installs.
When to use
- After Equip stage in book-learn-distill
- After editing
descriptionor trigger phrases in any Skill - Before publishing a Skill copy to a public repo
Test protocol
1 · Positive triggers (5 phrases → should activate)
| # | User says | Activated? |
|---|---|---|
| 1 | ☐ | |
| 2 | ☐ | |
| 3 | ☐ | |
| 4 | ☐ | |
| 5 | ☐ |
Use varied wording — not copy-paste of description.
2 · Negative triggers (3 adjacent tasks → should NOT activate)
| # | User says | Should route to | False positive? |
|---|---|---|---|
| 1 | ☐ | ||
| 2 | ☐ | ||
| 3 | ☐ |
3 · Self-check question
Ask Agent: "When would you use this skill?"
Compare answer to intended triggers in SKILL.md.
4 · Metadata audit
-
descriptioncontains clear trigger phrases - Negative constraints in SKILL.md body
- Boundary vs similar Skills documented
5 · Verdict
- Pass → record date in project
meta.json(skillTestedAt) if applicable - Fail → revise SKILL.md; log in references/template.md
Agent instructions
When user says "run skill activation test for X":
- Read
skills/<slug>/SKILL.md(or path given) - Propose 5 positive + 3 negative test phrases before user runs them
- After user reports results, output pass/fail table + fix list for description/triggers
Reference
Blank template: references/template.md