code-review

Review diffs, commits, and PRs to detect defects with severity-ranked findings.

1|Updated May 12, 2026
One-click install
npx skills add https://github.com/mystilleef/spae-framework --skill code-review-mystilleef
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/mystilleef/spae-framework/tree/main/skills/code-review
Command: npx skills add https://github.com/mystilleef/spae-framework --skill code-review-mystilleef

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Code changes often ship with hidden bugs, security flaws, or resource leaks that manual review misses. This Skill performs a skeptical, evidence-based review of diffs, commits, PRs, or changed files, flagging only material defects with exact code citations and severity rankings. ## Core Features & Use Cases - Multi-phase defect sweep: Scans eight categories (correctness, null safety, concurrency, security, resource management, error handling, API misuse, data integrity) across up to 7 phases with ledger-based deduplication. - Structured report output: Writes a schema-conformant code-review-report.yaml with findings sorted by severity (Critical to Nit) plus suppressed findings with documented rationale. - Verdict generation: Produces a clear verdict (Request Changes, Comment, or Approve) derived strictly from confirmed, exploitable findings. - Use Case: Before merging a pull request, invoke the review to catch an unguarded null dereference or an unsanitized input reaching a SQL query, with exact file and line citations. ## Quick Start Ask the agent to review the current uncommitted changes or a specific pull request for bugs and security issues.

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 pull request for bugs automatically?▼

Invoke the code review with the PR, commit, diff, or changed files as scope. It scans eight defect categories across multiple phases, then emits severity-ranked findings with exact path and line citations plus a verdict of Request Changes, Comment, or Approve.

What types of defects does automated code review detect?▼

It detects correctness errors, null dereferences, concurrency races, injection and hardcoded credentials, resource leaks, silently discarded errors, API misuse, and data integrity issues. Style, formatting, missing tests, and hypothetical inputs are explicitly out of scope.

Can I review only uncommitted local changes?▼

Yes. When no files or PR are specified, the review scopes itself to staged, unstaged, and untracked changes in the repository. It aborts with a clear message if no reviewable scope is detected.

Why are some potential issues missing from the review findings?▼

Findings lacking concrete exploit paths or carrying qualifying intent evidence are moved to suppressed_findings with a documented rationale. The review prioritizes precision over recall, dropping speculative or out-of-threat-model candidates.

What output format does the code review produce?▼

It writes a code-review-report.yaml file to the project root conforming to a JSON schema, containing scope, result, severity-sorted findings with excerpts and suggestions, and optional suppressed findings. A terse markdown summary with verdict is also emitted.