Review
Memory-safe engineering-peer plugin for Claude Code + Codex: 11 skills, 6 bounded agents, 4 hooks. Claims only what it can prove; closes every turn with one next action.
npx -y skills add menot-you/n-soul --skill 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
- 15 days oldThe repository was created 15 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.
- 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.
What its author says it does
Copied from the file, not written here
Perform an adversarial, read-only review of the actual candidate against its base, contracts, tests, safety, and rollback.
SKILL.md
1.9 KB, 375 tokens by cl100k_base, as published. Nobody here has run it
Review
Use for a change set, branch, pull request, plan, or release candidate. Default to read-only. Never repair findings unless the engineer separately asks for implementation.
Inputs to lock
- Exact candidate and base identifiers.
- Required result and acceptance criteria.
- Repository rules and affected trust boundaries.
- Claimed proof and rollback.
Method
- Pin the candidate and base. Read the actual diff or artifact, not a recap.
- Separate authored changes from generated or vendored bulk so meaningful logic is not hidden.
- Trace each changed input through callers, state transitions, external state changes, and user-visible output.
- Test six dimensions: correctness, reliability, maintainability, reversibility, security, and reach.
- Challenge every acceptance claim with the verification that can independently prove it.
- Look for partial failure, stale state, race conditions, idempotency failure, unsafe retry, privilege expansion, secret exposure, and rollback gaps.
- Run focused checks when read-only execution is safe. Never mutate the candidate to make review green.
- Rank findings by impact:
- Blocker — unsafe or fundamentally wrong; cannot ship.
- High — likely serious user or system failure.
- Medium — real defect with bounded impact.
- Low — maintainability or edge-case weakness worth fixing.
- Cite the smallest useful location and explain input, failure path, and impact.
- Return
LGTMonly when no blocking finding remains and required verification is green.
Output
Lead with findings, highest impact first. Then list open questions and proof gaps. If no findings remain, state LGTM and name the proof reviewed. Do not pad the result with a change recap, process theater, or menu of possible review modes.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.