code-review

Reviews a block's git diff adversarially for correctness bugs, edge cases, and unsatisfied acceptance criteria.

Updated Jun 12, 2026
One-click install
npx skills add https://github.com/lucolucus/mismagent --skill code-review-lucolucus
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/lucolucus/mismagent/tree/main/plugins/mismagent/skills/code-review
Command: npx skills add https://github.com/lucolucus/mismagent --skill code-review-lucolucus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Tests and structural checks can pass while code is still wrong: logic bugs hide behind passing tests, edge cases go unhandled, and acceptance criteria are satisfied only on paper. This Skill performs a semantic, adversarial review of a single block's diff in a fresh-context subagent, catching what deterministic verifiers cannot. ## Core Features & Use Cases - Three review lenses: Blind Hunter finds correctness bugs without trusting names or comments; Edge Case Hunter walks every branch, boundary, empty/error state, and concurrency path; Acceptance Auditor verifies every AC is truly satisfied, including dedicated audits of concurrency claims (real contention vs. sequential theater, atomicity vs. TOCTOU races). - Finding triage: Every finding is classified by severity (HIGH/MED/LOW) and disposition (Patch, Defer, or Decision), producing a strict handoff verdict of APPROVE, CHANGES, or BLOCKED. - Read-only by design: It finds and triages but never fixes, commits, or moves files; fixes belong to the worker. - Use Case: After the structural verifier passes a block in the mismAgent build movement, invoke this review on the block's diff to confirm the code is actually correct before merge. ## Quick Start Ask the worker-composer to run the code-review skill on the current block's diff after the verifier passes, and report the verdict with triaged findings.

Frequently Asked Questions about code-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a code diff for bugs that tests miss?

Run this review on the block's git diff in a fresh-context subagent. It applies three lenses: a Blind Hunter for correctness bugs, an Edge Case Hunter for boundary and error paths, and an Acceptance Auditor that checks each AC is genuinely satisfied.

What is the difference between the mismAgent verifier and code review?

The verifier is structural and deterministic: it checks build, tests, contracts, and ADR enforcement. The code review is semantic: it judges whether passing tests prove the right thing and whether ACs are satisfied in spirit, not just on paper.

How are concurrency acceptance criteria audited in code review?

Each concurrency-claim AC gets a dedicated audit checking whether the test creates real contention with threads and a start barrier, and whether the guarded operation is atomic rather than check-then-act. A sequential test covering a concurrency AC is reported as HIGH severity.

Can the code review skill fix the issues it finds?

No, it is strictly read-only. It finds and triages issues into Patch, Defer, or Decision dispositions; the worker applies Patch fixes, and Decision findings block the merge until a human decides.

When does a code review return CHANGES versus BLOCKED?

CHANGES is returned when at least one HIGH finding exists or an AC is not truly satisfied, triggering up to two worker fix cycles. BLOCKED is returned when a finding requires a human Decision that cannot be invented automatically.