AsyncReview tackles the limitations of traditional AI code review tools by employing Recursive Language Models (RLM) for autonomous, context-aware analysis of GitHub Pull Requests and Issues. It targets developers seeking deeper, more reliable code feedback, offering grounded insights beyond simple diff analysis.

How It Works

The core is an agentic, recursive loop: reason/plan, generate Python code, execute in a secure Python REPL sandbox. This sandbox allows LLM queries and tool commands, interacting with GitHub API via a tool interceptor (FETCH_FILE, SEARCH). The agent observes results, refines understanding, and repeats recursively. This contrasts with static, diff-only analysis by enabling repository exploration, context fetching, and verification, leading to grounded, cited feedback.

Quick Start & Requirements

No installation required; run directly using npx.

• Primary Command: npx asyncreview review --url <github_pr_or_issue_url> -q "<review_query>"
• Prerequisites: Node.js/npm (implied). Gemini API Key (GEMINI_API_KEY) for public repos. GitHub Token (GITHUB_TOKEN or --github-token flag) for private repos.
• Agent Integration: Add as a skill via npx skills add AsyncFuncAI/AsyncReview for vercel/skills compatible agents.

Highlighted Details

• Agentic Exploration: Autonomously explores repos, fetches context, and verifies findings.
• Full Repository Context: Analyzes beyond diffs, reading any file to understand dependencies.
• Grounded & Cited Feedback: Provides answers grounded in file paths/lines, minimizing hallucinations.
• Recursive Reasoning Loop: Multi-step planning, coding, execution, and observation for iterative refinement.
• Extensible Skill: Designed for integration with other agentic systems.

Maintenance & Community

No specific details regarding contributors, sponsorships, or community channels were provided in the README snippet.

Licensing & Compatibility

• License: MIT.
• Compatibility: MIT permits broad usage, including commercial applications and closed-source linking.

Limitations & Caveats

Requires specific API keys (Gemini, GitHub), especially for private repos. Primarily presented as an integrated 'skill' for agentic frameworks, suggesting optimal use within existing agent setups rather than standalone deployment for non-agent users. No explicit alpha status or known bugs were mentioned.