Skill Trigger-Test Protocol
The five-minute test for whether a skill actually fires: paraphrase matrix, fire-rate scoring, and description fixes. Copy it below, fill the bracketed fields, delete what you don't need.
# Trigger test: [skill name]
Date: [date] · Skill description under test: "[paste frontmatter description]"
## Paraphrase matrix
Write 3 phrasings a real user would use for a task this skill should handle,
and 2 for adjacent tasks it should NOT handle.
| # | Prompt | Should fire? | Fired? |
|---|--------|--------------|--------|
| 1 | [natural phrasing] | yes | [y/n] |
| 2 | [different vocabulary] | yes | [y/n] |
| 3 | [indirect/implicit phrasing] | yes | [y/n] |
| 4 | [adjacent task, out of scope] | no | [y/n] |
| 5 | [superficially similar, excluded] | no | [y/n] |
## Scoring
- Fire rate on should-fire: [n/3] (pass = 3/3)
- False fires on should-not: [n/2] (pass = 0/2)
## If it failed
- Silent non-use (rows 1-3): rewrite the description as task-shaped trigger conditions; add the failing prompt's vocabulary
- False fires (rows 4-5): add explicit exclusions ("Do NOT use for...")
- Re-run the full matrix after every description change
Field notes
| Field | Why it matters |
|---|---|
| Fresh sessions | Each row runs in a fresh session; triggering carries over within a session and contaminates the test. |
| Should-NOT rows | Half the matrix tests exclusions because misfires erode trust as fast as non-use. Most published skills have never been tested on either. |