Passed 21 of 23 test cases with the skill, and 12 of 23 without it.
Without skill52%
With skill91%
Checks it passed that the model alone missed
Prohibits merging superseded authority lines.
Evaluates RDD switch status before issuing review receipts.
Requires creating a new candidate when validation findings occur.
and 14 more in the full report
10 fixed11 passed either way1 still failing1 broke
Pass-rate lift
+39 pp
Tokens
+13%
Agent turns
0%
Response time
-7%
Model gemini-3.6-flashMeasured 9/29/2026v121 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 10
Prompt
When drafting the operator flow inventory for CLI issue #61, the contributor documented entry points, execution modes, environments, and expected positive outputs. What required inventory component is missing?
What was checked, without and with the skill
Checked
Without skill
With skill
Requires including negative controls in every operator flow inventory.
Missed
Passed
case-21 · gemini-3.6-flash · 9/29/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md45 linessingle file
Activation Contract
Load when the frontmatter trigger terms apply to a defect workflow.
This skill guides public collaboration. It does not grant issue approval, label, review, exception, or merge authority.
Hard Rules
Check the user-owned RDD kill switch first. When disabled, do not start receipt reviews or fabricate approval; follow ordinary policy and report disabled/unmanaged.
Require an approved issue (status:approved) and clean current main reproduction before implementation. Audit existing PRs for supersession or conflict; stop or narrow stale claims.
Group by causal authority invariant. Use one issue and one PR or explicit chain per independent invariant and rollback boundary. Split independent causes; never merge a superseded or conflicting authority line.
Inventory every operator flow claimed by the issue or PR, including entry, mode, environment, expectation, and negative controls. Require one truthful black-box bench journey per CLI or lifecycle flow, or actual runtime E2E proof when the core bench cannot represent it. Synthetic proxy coverage never proves another runtime.
Use CodeGraph-first impact mapping, a dedicated worktree, and behavior-first tests. Run source-mutating normalization before candidate freeze.
Forecast authored changes before edits. The hard limit is 400 additions plus deletions; above it, STOP for a chain or explicit maintainer-approved exception.
Only when RDD is enabled, bind the review candidate identity, lineage, correction, and recovery records exactly. Keep bounded review defects in one correction transaction; ordinary repository policy decides delivery.
Require independent read-only candidate validation before publication. Validation cannot edit source or authority; findings require a new candidate.
Keep communication humane and evidence-based. Repository labels and workflow metadata are maintainer-owned, never evidence of contributor blame.
Decision Gates
Condition
Action
RDD disabled
Ordinary policy; disabled/unmanaged; no receipt or approval claim.
Issue gate or reproduction fails
Wait, stop, or narrow with evidence.
Invariant or rollback is independent
Separate issue and authoritative PR line.
Core bench fits / does not fit
Bench journey / actual runtime E2E; never proxy.
Forecast exceeds 400 lines
Chain or approved exception before edits.
Execution Steps
Check mode, approval, PR conflicts, and current-main reproduction.
Name invariant and rollback; isolate the worktree; CodeGraph-map code, tests, evidence, docs, distribution, and registration.
Inventory flows and controls; add failing tests and the smallest correction.
Normalize, enforce budget, run tests, and record each flow's exact candidate, command, scenario, and result.
Freeze, validate read-only, and give the verdict, evidence, and one humane next action.