Passed 8 of 18 test cases with the skill, and 4 of 18 without it.
Without skill22%
With skill44%
Checks it passed that the model alone missed
Requires fixing the calculation bug in `checkout.py`.
Reads the repository check recipes before executing check commands.
Assigns `subject-manifest.v1` derivation to the integrating caller.
and 5 more in the full report
5 fixed3 passed either way9 still failing1 broke4 errored
Pass-rate lift
+22 pp
Tokens
+70%
Agent turns
0%
Response time
-41%
Model gemini-3.6-flashMeasured 10/10/2026v8 · latest is v98 of 22 cases passed with skill · 18 comparable
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
See it in action
One of 5 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 5
Prompt
We need to run code quality checks for `pkg/crypto`. Rather than looking up the repo's Makefile, can we run standard `go test ./...` from memory?
What was checked, without and with the skill
Checked
Without skill
With skill
Reads the repository check recipes before executing check commands.
Missed
Passed
Rejects relying on remembered standard CLI invocations when project recipes exist.
Missed
Passed
case-12 · gemini-3.6-flash · 10/10/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md137 lines5 filesincludes scripts
Implement
Implement the accepted outcome. Repair ordinary known defects directly. Use the existing
intent; no Plan, Recall or Learn worksheet is owed for a clear edit. Implement
owns source changes and factual checks; the runtime derives identity and receipts.
Rules that decide the result
Fix the cause, not the oracle. Never loosen an assertion, tolerance,
golden, fixture or suppression, or substitute a mock or placeholder, to turn
a check green. A check that fails against accepted behavior points at a
product defect; changing what the check accepts is an acceptance change and
needs caller authority.
Find live consumers before editing. Search the callers, readers, scripts,
docs and tests of every edited function, type, path, format or message. For
each one, state whether the change affects it and which check covers it. When
retiring or renaming, consumers include installations, lookups and old-name
invocations; preserve historical provenance.
Name discriminating checks. A behavioral change preserves RED
for the expected missing behavior: name the check that fails before the edit
and passes after, plus the check for each affected consumer.
Report only what ran. Give exact commands and results. List every check
you did not or could not run as not run. Never call unrun work verified,
fixed or green.
For authorized service operations, load only the relevant procedure from
operations methods (reliability, delivery, incident
recovery, resilience, toil); ordinary edits owe no operations phase.
Workflow
Read intent, acceptance, scope and repository boundaries before the first
write; reuse loaded contracts. RPI boundaries
apply when that workflow is explicitly selected. Caller-selected episode
tracking follows the session association reference;
keep unknowns explicit and invent no parentage or second tracker.
Carry the accepted behavior examples forward unchanged. Use repository
domain names in symbols and tests; check observable outcomes through the
relevant interface. Run the smallest applicable check before and after
editing. Pure refactors, relocations or docs may have an honest green