Harness review
Review code, documentation, commits, or pull requests read-only and report evidence-backed actionable findings classified P0-P3. Use when a user asks for a review, audit of a diff, risk assessment, or validation of proposed changes; apply fixes only when separately requested.From its SKILL.md
npx -y skills add cekrauseee/harness --skill harness-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.2 KB, 371 tokens by cl100k_base, as published. Nobody here has run it
Harness Review
Review for concrete defects and risks. Reviews are read-only by default: identify actionable findings without changing files, commits, branches, pull requests, or external state.
Review Procedure
-
Establish the requested scope and applicable repository rules.
-
Inspect the complete relevant diff and enough surrounding code, tests, configuration, or documentation to understand behavior.
-
Verify each suspected issue. Trace inputs, control flow, affected callers, and existing tests instead of relying on pattern matching alone.
-
Classify actionable findings with references/review-severity.md.
-
Keep line ranges tight and ensure the cited line is part of, or directly explains, the defect.
-
Optionally validate structured findings before rendering them:
python3 scripts/validate_review.py findings.json -
Present findings in priority order. If none remain after verification, state that no actionable findings were found and mention any material verification gap.
Finding Contract
Each finding contains:
[P2] Handle the empty-token path
path/to/file.py:42
Evidence: ...
Impact: ...
Direction: ...
- Use an imperative, concise title without a terminal period.
- Explain observable evidence or a reproducible path.
- State the concrete impact, not a vague quality concern.
- Suggest direction without prescribing a large unrelated rewrite.
- Do not report praise, summaries, stylistic preference, speculative edge cases, or issues already prevented elsewhere.
Mutation Boundary
- A review request does not authorize fixes, formatting, commits, comments, approvals, or requested-changes state.
- Apply fixes only when the user separately requests implementation.
- Publish review comments or submit a platform review only when explicitly authorized.
- Do not weaken or inflate a priority to influence workflow.
What ships with it: 3 files
4.8 KB alongside SKILL.md, 1 of them executable
agents/
- openai.yaml275 B
references/
- review-severity.md1.3 KB
scripts/
- validate_review.pyruns3.2 KB