gds-code-review

Reviews code changes with parallel adversarial review layers and structured triage.

Updated Sep 3, 2026
One-click install
npx skills add https://github.com/watchthelight/shatterfish --skill gds-code-review-watchthelight
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gds-code-review
Source: https://github.com/watchthelight/shatterfish/tree/main/.claude/skills/gds-code-review
Command: npx skills add https://github.com/watchthelight/shatterfish --skill gds-code-review-watchthelight

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Code reviews often miss edge cases or produce noisy, unstructured feedback. This Skill runs adversarial multi-layer reviews of code diffs and triages every finding into clear actionable categories, so nothing slips through and nothing wastes your time. ## Core Features & Use Cases - Parallel Adversarial Review Layers: Launches 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 defects. - Structured Triage: Normalizes, deduplicates, and classifies findings into decision_needed, patch, defer, or dismiss buckets, then writes them into the story file. - Sprint Integration: Updates story status and syncs sprint-status.yaml after the review completes. - Use Case: After finishing a feature branch, ask for a code review; the Skill gathers the diff, runs the three review layers, presents triaged findings, applies approved patches, and marks the story done. ## Quick Start Ask the assistant to run a code review on your current branch changes against main.

Frequently Asked Questions about gds-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 my changes?

Trigger the review by saying "run code review" or "review this code". The Skill identifies the diff source from your message, conversation, sprint status, or git state, then launches parallel review layers and triages the findings.

What diff sources can a code review workflow accept?

It accepts uncommitted changes, staged changes only, a branch diff against a base branch, a specific commit range, a user-provided diff, or a file list. The Skill validates each source and halts if the diff is empty or unparseable.

How are code review findings classified and handled?

Findings are deduplicated and classified as decision_needed, patch, defer, or dismiss. Decision items require your input, patches can be applied automatically, deferred items go to a deferred-work log, and dismissed noise is dropped.

Can code review work without a spec or story file?

Yes, it runs in no-spec mode where the Acceptance Auditor layer is skipped. Findings are still reviewed and triaged, but nothing is written to a story file and sprint status is not synced.

What happens if a review subagent fails or returns nothing?

The failed layer name is recorded and the review continues with the remaining layers. If zero findings remain and any layer failed, the Skill warns that the review may be incomplete instead of declaring it clean.