MergeWarden is a specialized change-control gate designed specifically for pull requests generated by AI coding agents. As AI coding agents increasingly automate development tasks and open PRs throughout the day, MergeWarden ensures these agents stay strictly within predefined repository boundaries.

Built following an analysis of 2,204 merged agent PRs—which revealed that zero agents declared their scope—MergeWarden operates as a checkout-free GitHub Action and CLI. It screens pull requests against declared scopes, agent-control-plane drift (including files like AGENTS.md and mcp.json), and potential workflow permission escalations.

Crucially, MergeWarden requires no code checkout and no LLM, delivering deterministic, replayable evidence. Developers can instantly test what the tool catches with a single command without needing tokens or a repository using npx mergewarden demo. Released under an MIT license, it provides essential governance for AI-driven software development workflows.

Pros of MergeWarden

Acts as a dedicated change-control gate for AI-agent pull requests
Operates completely checkout-free as a GitHub Action and CLI
Uses no LLMs, ensuring deterministic and replayable evidence
Quickly tests functionality locally with a simple command (npx mergewarden demo)

Cons of MergeWarden

Requires proper configuration of repository boundaries and control-plane files
Focuses strictly on agent boundaries and permissions rather than general code quality review

Frequently Asked Questions

What does MergeWarden do?

MergeWarden acts as a change-control gate for AI-agent pull requests, checking whether agents stayed within repo boundaries such as declared scope, agent-control-plane drift (AGENTS.md, mcp.json), and workflow permission escalation.

Does MergeWarden use an LLM?

No, MergeWarden does not use an LLM. It relies on deterministic evaluation to provide replayable evidence without involving artificial intelligence processing during checks.