code-review

Reviews code diffs adversarially using parallel review layers with structured triage and patching.

3|Updated Feb 26, 2026
One-click install
npx skills add https://github.com/stellar-zk/stellar-zk --skill code-review-stellar-zk
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/stellar-zk/stellar-zk/tree/main/.claude/skills/code-review
Command: npx skills add https://github.com/stellar-zk/stellar-zk --skill code-review-stellar-zk

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews often miss subtle bugs, edge cases, and deviations from the spec because a single reviewer pass lacks diverse perspectives. This Skill runs adversarial multi-layer reviews over a git diff and turns raw findings into triaged, actionable outcomes. ## Core Features & Use Cases - Parallel Adversarial Review Layers: Runs a Blind Hunter (diff only), an Edge Case Hunter (diff plus project access), and an Acceptance Auditor (diff plus spec) to catch different classes of issues. - Structured Triage: Normalizes, deduplicates, and classifies findings into decision-needed, patch, defer, or dismiss buckets. - Actionable Outcomes: Writes findings into the story file, applies patches on request, tracks deferred work, and syncs sprint-status.yaml. - Use Case: After finishing a story branch, ask for a review of the branch diff against main; the Skill gathers the diff, runs the three review layers, presents triaged findings, and applies approved fixes. ## Quick Start Ask the assistant to run code review on your current branch diff against main and walk you through the findings.

Frequently Asked Questions about code-review

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

FAQPage Schema
How do I run an adversarial code review on a git diff?▼

Provide a diff source such as a branch, commit range, staged changes, or a pasted diff. The Skill constructs the diff, launches parallel review layers (Blind Hunter, Edge Case Hunter, Acceptance Auditor), then triages and presents the findings.

What diff sources does this code review workflow support?▼

It supports uncommitted changes, staged-only changes, branch diffs against a base branch, specific commit ranges, user-provided diffs, and explicit file lists. It validates each source and halts if the diff is empty or unparseable.

Can code review work without a spec or story file?▼

Yes. Without a spec it runs in no-spec mode, skipping the Acceptance Auditor layer. Findings that would need a decision are reclassified as patch or defer, and nothing is written to a story file.

What happens if a review subagent fails or returns nothing?▼

The failed layer is recorded in a failed_layers list and the review continues with the remaining layers. If no findings remain and layers failed, the user is warned the review may be incomplete rather than told it is clean.

How are review findings classified and handled?▼

Findings are deduplicated and classified as decision-needed, patch, defer, or dismiss. Decision-needed items are resolved with the user first, patches can be applied automatically, and deferred items are logged to a deferred-work file.

Does the review update sprint status after completion?▼

Yes, when a story file and story key exist. The story status is set to done if all findings were resolved, or in-progress otherwise, and the matching entry in sprint-status.yaml is updated while preserving its structure.