asyncfuncai/asyncreview

Overview

This skill provides AI-powered reviews of GitHub pull requests and issues with agentic access to the codebase. It helps you analyze diffs, ask targeted questions about changes, and get step-by-step reasoning and evidence-backed findings. The tool supports public and private repos (with token handling) and offers expert-level review modes for deeper analysis.

How this skill works

The agent extracts the PR/issue URL, checks repository visibility, and ensures required credentials (GEMINI_API_KEY and optional GITHUB_TOKEN) are available. It fetches changed files and can autonomously retrieve additional files outside the diff to reproduce context in a sandbox. The review runs a sequence of reasoning steps, code inspections, and optional expert checks, then returns findings with sources and suggested fixes.

When to use it

• You need an AI review of a GitHub pull request or issue.
• You want to verify whether a PR breaks existing callers or functionality.
• You need focused questions answered about code changes (e.g., security, API compatibility).
• You want an expert audit covering SOLID, performance, and security concerns.
• You need evidence-backed findings with file-level sources for triage or code review.

Best practices

• Ensure GEMINI_API_KEY is set before running any review.
• Check repo visibility early; set GITHUB_TOKEN for private repos or higher API quotas.
• Provide a clear, specific question for targeted reviews; use --expert for full audits.
• Use URL formats https://github.com/owner/repo/pull/123 or /issues/456 and include the exact PR/issue link.
• Choose output format (pretty, markdown, json) that fits your workflow; use JSON for automation.

Example use cases

• Run a quick review asking: “Does this change break any existing callers?” and get file-level evidence.
• Audit a PR for security vulnerabilities using: npx asyncreview review --url <PR_URL> -q "Does this PR introduce any security vulnerabilities?"
• Check whether a deprecated function is still used by asking the agent to fetch src/utils.py and search usages.
• Perform an expert code review to surface SOLID violations, performance hotspots, and P0/P1 fixes.
• Integrate JSON output into CI to fail builds on critical findings or to generate reviewer TODOs automatically.