Code review
Portable equal-primary Claude and Codex agent harness for a governed agentic SDLC
npx -y skills add mblauberg/provenant --skill code-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 27 days oldThe repository was created 27 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 2 stars2 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.
What its author says it does
Copied from the file, not written here
Use for source-read-only review of a PR, diff, commit, migration, refactor, or implementation through its dependency cone. Not for root-cause diagnosis, authorised fixes, or browser/UX review; use diagnose, implement, or ui-ux-design.
SKILL.md
3.6 KB, as published. Nobody here has run it
Code review
Review is source-read-only unless the user separately authorises fixes. A reviewer may write compressed findings and traces only to an assigned artifact directory. The diff is the entry point, not the boundary. Judge the resulting system, not the changed lines alone.
Review
- Read project instructions and preserve the requested scope and output contract.
- Establish the change intent from the spec, issue, tests, commit message, or user request. Mark unclear intent instead of inventing it.
- Trace the affected dependency cone: full touched files, live callers/
consumers, exports, canonical owners, tests, configuration, migrations,
persistence boundaries, dependency/lock changes, and generated artefacts.
Record inspected/excluded surfaces; unavailable context stays
unknown. - Select review lenses from the task/risk profile. Correctness/spec alignment is mandatory; add security/privacy, data/concurrency, performance/reliability, tests, architecture, readability, UX/accessibility or operations only when activated by the dependency cone.
- For substantial+ or wide-surface work, use independent targeted agents with
distinct primary lenses and deliberate overlap on the riskiest invariant.
orchestrateadds the required other-primary and distinct-family review pressure fromHARNESS.md. Then run an anonymised claim challenge and a fresh reduction. Never rank prose or majority-vote findings. - Review the design delta. Look for duplicated ownership, scattered flags, nullable modes, casts hiding invariants, thin wrappers, parallel flows, and abstractions adding concepts without reducing complexity.
- Ask whether a proven reframe could delete branches, state, layers, or duplicated flows. Size is a signal, not a defect. Past 1000 lines a source file is a finding; tests only warn.
- Check tests and verification against the acceptance criteria. Confirm the trajectory: relevant deterministic checks ran and their results are available. Never infer coverage from a green summary alone.
- Report only high-confidence, actionable findings. Return
cleanwhen no genuine defect exists.
Load review-lenses.md for the detailed inspection questions, multi-agent-review.md for lens fan-out/council reduction, and finding-contract.md before writing findings.
Boundaries
orchestratechooses reviewer topology, families, and waves. This skill defines what each reviewer inspects and the independence/reduction contract.diagnoseowns reproduction and root cause for known broken behaviour.implementowns authorised fixes and bounded re-review;tddorrefactormay supply the method. Do not mutate source during review.- Artifact-only authority permits named outputs under the assigned run directory; it does not permit arbitrary repo-root scratch. Use the system temporary directory when no run directory exists. Never redirect a command over a wildcard/list that can include its own growing output; keep captures bounded.
- Language, framework, UI, security, and project skills add specialised lenses; do not repeat their doctrine here.
- A structural alternative blocks only when tied to a present defect or material regression with a safer, validated design. Attractive redesign alone is not a finding.