Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsSkill Trigger Test: Should-Fire and Should-Not-Fire Cases
Write should-fire and should-not-fire prompts for one skill, record what each run showed, and count hits, false fires and cases nobody saw. Runs no model; you record the outcome.
$5
Skill Trigger Test: Should-Fire and Should-Not-Fire Cases
Example session with this skill installed
Build a trigger test set for my csv-tidy skill, check it, and score the results I recorded from sixteen fresh sessions.
- Read your context and instructions
- Compiled the trigger test
Summary
The agent checked the skill's name and description against the Agent Skills specification, checked a 16-case set (8 should fire, 8 should not, 3 near misses), scored the recorded results and compared the description's wording with two other installed skills. The results in this demo are the example data that ships with the skill, not a real run.
$ python3 scripts/frontmatter_check.py references/example/csv-tidy/SKILL.example.md
OK name-length: 8 characters (source: F1 "Must be 1-64 characters")
OK name-matches-folder: name equals the folder name 'csv-tidy' (source: F1 "Must match the parent directory name")
OK description-length: 304 characters (source: F1 "Must be 1-1024 characters")
FRONTMATTER ok=9 fix=0 warn=0 unknown=0 result=ok
$ python3 scripts/make_suite.py --check references/example/suite.json
SUITE-CHECK cases=16 fix=0 warn=0 unknown=0 result=ok
$ python3 scripts/score.py references/example/suite.json references/example/results.json
run (as recorded; the result holds only for this run): agent=example agent (made-up data, not a real run), model=example-model, date=2026-01-15, other_skills=json-format, pdf-merge
| F01 | should_fire | | fired | hit | Clean up this CSV, the header row is repeated three times. | fresh session; the transcript showed csv-tidy loaded befo... |
| N03 | should_not_fire | yes | not fired | no_fire | Pretty print this JSON file with sorted keys. | fresh session; json-format loaded, csv-tidy not loaded |
near_miss false_fire=0/3
TRIGGER-TEST hit=8/8 false_fire=0/8 unknown=0 result=pass
$ python3 scripts/overlap.py references/example/csv-tidy/SKILL.example.md --peers references/example/peers
OK json-format/SKILL.example.md: 0 shared phrases (source: heuristic)
OVERLAP peers=2 with_shared=0 unknown=0 result=ok
| Check | Result |
|---|---|
| Name and description against the spec | 9 OK, 0 to fix |
| Case set | 16 cases, 0 to fix |
| Trigger score | 8 of 8 hits, 0 of 8 false fires, 0 unknown |
| Shared wording with other skills | 0 shared phrases |
Next steps
- Run each prompt in a fresh session and record true, false or null with what you saw.
- Score again; any null or missing evidence counts as unknown, and a pass needs 0 unknown.
- Remember that a pass holds only for the agent, model and installed skills you recorded.
Connects securely to your tools. The creator never sees your data.
About this skill
Use when you want to know whether your skill is loaded on the prompts it is meant for and stays out of the ones it is not, when it keeps loading on a nearby task, or before you change its description.
It builds a case set with at least 8 should-fire and 8 should-not-fire prompts, 3 of them near misses, and checks the filled set for empty and duplicate prompts. You or the agent run each prompt and record true, false or null with the evidence you saw; the score counts a case with null or no evidence as unknown, never as passed, and a pass needs 0 unknown.
It also checks name and description against the Agent Skills specification with a quote per rule, and lists 2 and 3 word phrases the description shares with your other skills (heuristic).
Do not use when you want the description rewritten, a whole package checked before upload, or the quality of the skill's work judged once it is loaded.
Requires: Python 3.9+ (standard library only). The scripts read and write local JSON and Markdown files only; they run no model, start no process and make no network calls.
Independent: Independent tool. Not affiliated with or endorsed by Anthropic. Also not affiliated with OpenAI, Anysphere (Cursor) or Google. Product names are used only to describe compatibility.
It does not run prompts and cannot see which skill an agent loads; it counts what you recorded. A pass holds only for the runs you recorded, with that agent, model and set of installed skills. No claim about how a particular agent chooses skills or that a skill will be loaded. It does not judge what a skill does once it is loaded. The phrase overlap check compares wording only and is a heuristic.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 4 days ago
Requires
Python