Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →notque/systematic-code-review
4-phase code review: UNDERSTAND, VERIFY, ASSESS risks, DOCUMENT findings.
Passed 20 of 21 test cases with the skill, and 11 of 21 without it.
and 7 more in the full report
10 fixed11 passed either way1 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
During review, a candidate finding claimed `UserContext` could be null, causing a crash. However, caller tracing in Phase 1 proved that the upstream middleware guarantees non-null context before calling this function. What is the disposition of this candidate finding?
| Checked | Without skill | With skill |
|---|---|---|
| Marks the candidate finding as REFUTED due to being already handled or hypothetical | Passed | Passed |
| Excludes the refuted finding from the final verdict findings list in Phase 4 | Missed | Passed |
The instructions your model reads when the skill loads.
Review the change, verify its claims, assess risk, and report actionable findings. This is the shared review procedure. Parallel review divides its scope among reviewers; it does not add another full review afterward.
| Task | Reference |
|---|---|
| Select PR review scope and roster | ../../process/pr-workflow/references/pr-risk-policy.md |
| Review Go exports, concurrency, resources, metrics, or tests | references/go-review-patterns.md |
| Classify a finding | references/severity-classification.md |
| Respond to feedback | references/receiving-feedback.md |
Read applicable repository instructions and the complete diff. Read enough surrounding code to understand each changed path and its consumers; load whole files when their structure matters. Check the requested outcome, compatibility requirements, and affected dependencies.
When signatures, parameter meanings, or sentinel values change, find all callers and interface implementations. Search receiver syntax such as .GetEvents(; use type-aware references such as gopls when available. Trace each parameter to its source:
"*".Check validation at each caller. A reachable user input that bypasses a security filter is blocking. Verify the caller set yourself; the PR description may omit callers.
Record the reviewed base/head or working diff, scope, and material unknowns. Ask only when missing information prevents a sound review; continue independent checks.
Gate: Every changed path is accounted for, with affected callers traced where needed.
Reuse observed test and review results when they cover the current code, dependencies, configuration, and environment. Record their source and scope. A prior verdict alone is not evidence. After a fix, check the changed paths and affected consumers; repeat broader checks only when the fix invalidates their evidence or repository policy requires it.
Run relevant tests for uncovered behavior and all required repository checks. Inspect actual output and retain logs; report commands, results, and limitations without pasting full logs. Missing tools, skipped checks, and inferred outcomes are not passes. Identify whether a test failure is caused by this change or is unrelated; required failing checks still prevent merge.
Check material claims in comments and the PR description against code, callers, tests, or observed behavior. Verify edge cases and coverage of changed paths. Source inspection can establish structure, but cannot substitute for a runtime result when that is the claim.
Gate: Claims have supporting evidence; required checks passed or the review records the gap without approving it away.
test cases passed, vs 11 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.