Skill Test Runner (Trigger and A/B)
Test a skill in fresh sessions: ready-to-run prompts, commands, and rubric for trigger recall, false triggers, and blind A/B.
Before & after
Same prompt, same agent — the only difference is the skill.
Running a skill on a task generally depends on the platform you're using. In most setups: - The skill activates when your request matches what it's designed for - You describe your task and the skill guides the response - Some skills need specific inputs or context If you tell me which skill and what you're trying to do, I can help you figure out how to invoke it. Different skill collections have different conventions for triggering and using them. What are you trying to run?
Here's the run workflow — a repeatable way to invoke a skill on a task
and get consistent output, rather than hoping it triggers.
THE FLOW
1. STATE THE TASK, matched to the skill
Describe what you want in terms the skill recognizes. A skill
activates on matching intent — "review this PR" hits the review
skill; "make it better" hits nothing in particular.
2. SUPPLY THE INPUTS IT NEEDS
Most skills expect specific context (the file, the transcript, the
brand voice). Give it up front; a skill starved of input produces
generic output, which reads as the skill not working.
3. LET THE SKILL'S WORKFLOW RUN
The skill applies its own instructions and validation — you don't
re-specify the steps each time. That's the point of a skill vs. a
one-off prompt: the process is baked in and repeats.
4. ITERATE WITHIN IT
Refine by adjusting inputs, not by re-explaining the whole task.
WHY IT STAYS CONSISTENT
The skill carries its own validation, so output holds its shape run to
run — the same task next week produces the same kind of result, because
the workflow, not your phrasing, drives it.
THE COMMON MISTAKE
Under-specifying the input and blaming the skill. Give it the file, the
context, the goal — then let its workflow do the rest.
Tell me the skill and the task, and I'll show you the exact invocation.About this skill
A skill that reads well can still fail in use. This runner tests it the way it will be used: realistic prompts in fresh sessions, measuring whether it triggers, whether it fires on the wrong requests, and whether its output actually beats no skill.
What it does
- Drafts should-trigger, near-miss, and edge-case prompts in real user phrasing
- Gives the exact commands to run each prompt in isolated fresh sessions, with and without the skill (or runs them when you ask)
- Grades outputs blind against an agreed rubric and checks known skill failure modes
- Maps misses and ties to specific description or body fixes, then re-tests
- Checks your own evidence: same-session or single-run "it works" results do not count
What you get
A ready-to-run test plan in the first reply, then a test report from the results: trigger recall and false triggers, A/B wins, losses, and ties with evidence, failure modes seen, and exact fixes.
Best for
Skill authors who want evidence a skill works before publishing or selling it.
SKILL.md preview
---
name: skill-test-run
description: Test a Claude Code skill by actually running it - a small set of should-trigger, should-not-trigger, and edge-case prompts in fresh sessions, checking whether it triggers, whether the output follows the skill, and whether it beats a no-skill baseline in a blind comparison - then report pass rates and the fixes. Use when the user says "test my skill", "does this skill actually trigger?", "run my skill against some prompts", "is my skill better than no skill?", "A/B test this skill", "eval this SKILL.md", or "why does it fire on the wrong requests?".
---
# Skill Test Run
… (sign up to view the full skill)More ai & agents skills
View all AI & Agents skills →New Skill Scaffolder
Create a new Claude Code skill correctly: right mechanism, valid name and folder, a triggering description, and a smoke test.
Skill Library Lint and Overlap Report
Lint every skill in a repo at once: frontmatter, name-folder match, weak descriptions, overlapping triggers, broken refs, plus a CI script.
Self-Verifying Agent Loop: Plan, Check, Retry, Close
Turns a goal into a task plan with machine-run checks, executes with retry caps, escalates, and refuses to close until verified.
Chief AI Officer Plan Review
Evidence-first review of one AI plan: eval gaps, cost per successful task, model and vendor risk, prompt injection, and EU AI Act screen.