Audit work, verify claims with concrete evidence, and explain the result in simple grade-5 language.
SkillScore64Maintained
Automatic
◆Coding pack
Last verified 8/10/2026
What it changes
Passed 14 of 22 test cases with the skill, and 8 of 22 without it.
Without skill36%
With skill64%
Checks it passed that the model alone missed
Outputs a functional Bash CI/CD script.
Ends with at most one specific next test recommendation.
Defines 'metadata' in plain terms as small facts about a file.
and 8 more in the full report
7 fixed7 passed either way7 still failing1 broke
Pass-rate lift
+27 pp
Tokens
+11%
Agent turns
0%
Response time
-23%
Model gemini-3.6-flashMeasured 8/10/2026v114 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
Audit this single bug fix in `utils/math.ts`:
```ts
export function add(a: number, b: number) { return a + b; }
```
We ran `jest utils/math.test.ts` and all tests passed. Explain what changed, why it matters, how it was verified, and what to do next.
What was checked, without and with the skill
Checked
Without skill
With skill
Includes the test runner command `jest utils/math.test.ts` as the verification evidence.
Passed
Passed
Ends with at most one specific next test recommendation.
Missed
Passed
case-12 · gemini-3.6-flash · 8/10/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md99 linessingle file
Audit, Verify, Explain
Core Rule
Treat every answer as three jobs:
Audit what changed or what is being claimed.
Verify it with direct evidence.
Explain it like the reader is smart but new to the topic.
Do not skip verification when local files, commands, logs, tests, screenshots, or source data are available. Do not pretend something was verified if it was only inferred.
Workflow
1. Audit
Start by finding the real source of truth:
For code changes, inspect the diff, touched files, related call sites, and existing tests.
For bug fixes, identify the before/after behavior and the user-facing path.
For performance claims, separate measured evidence from likely improvement.
For release or app behavior, check the packaged/running artifact when possible.
For documents or content, compare the user request against the actual produced artifact.
Look for:
obvious bugs or regressions
missing edge cases
stale assumptions
unverified claims
mismatches between implementation and user intent
risks that a grade-5 explanation might accidentally hide
2. Verify
Prefer evidence in this order:
Automated tests, builds, linters, typechecks, or validators.
Running the actual app or workflow.
Logs, process checks, screenshots, generated artifacts, or live output.