Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →jgravelle/mechanism-not-instance
The two questions that separate a fix for the reported spelling from a fix for the property. Load in /fix-issue steps 4-5 and in review of any fix.
Passed 17 of 25 test cases with the skill, and 3 of 25 without it.
and 24 more in the full report
14 fixed3 passed either way8 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.
We found a bug in how version strings are parsed at a specific API endpoint call site. Please help me structure our fix investigation notes—I need to analyze where this logic truly belongs, identify any variation forms of the input, and ensure our test plan actually validates the defect.
| Checked | Without skill | With skill |
|---|---|---|
| Directs recording the investigation notes in ISSUE.md. | Missed | Passed |
| Recommends moving the version string parsing fix one layer down into the core version authority. | Missed | Passed |
| Recommends using find_references to search for alternative input spellings. | Missed | Passed |
The instructions your model reads when the skill loads.
Authority: CLAUDE.md "Standing lessons" 08-19 (#506-#509: a second
generator, a second call site, a second derivation), 09-01 (#566: a guard
written against a spelling), 09-02 (#572: fixed in the cache, not at the
two call sites); docs/harness/ARCHAEOLOGY.md for the guard that already
names the lesson.
Ask, and write the answers into ISSUE.md:
find_references on the
fixed symbol and its callers; the qualified form, the aliased form, the
second transport, the other OS's separator. A test per spelling found.Also: run the reintroduced defect against the new test (Practice 9, non-vacuity). A guard green only against the fixed tree may be measuring nothing. And when an OLD test turns red, check whether it was the defect's witness before changing the code back.
test cases passed, vs 3 / 25 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.