MergeWarden

MergeWarden is an AI PR-governance tool that gates AI-agent pull requests against a repository's declared boundaries for teams shipping code with coding agents. It runs as a checkout-free GitHub Action, a CLI (npx mergewarden scan), and a hosted web queue, and its analysis is deterministic: it never executes pull-request code and never calls a language model, so its findings are replayable from the base commit. MergeWarden's detailed risk scan checks four security boundaries: workflow permission escalation (write-access gains or dangerous triggers), agent-control-plane drift (edits to AGENTS.md, CLAUDE.md, and .mcp.json), untrusted prompt inputs that could reach agent prompts, and install-script additions in package manifests. It also enforces a machine-checkable scope contract, declaring the agent, task, and allowed_paths, so a reviewer can see when an agent's PR strays outside what it said it would change. For triage, it surfaces queue-ordering facts such as PRs with no linked issue, thin descriptions under 80 characters, an unused PR template, or oversized diffs above 50 files or 1,500 changed lines. MergeWarden is MIT-licensed and open source. It is aimed at maintainers and platform teams who merge a high volume of agent-authored PRs and need deterministic, auditable evidence rather than another AI reviewer.

Best for maintainers and platform teams who need to gate high volumes of AI-agent pull requests against declared scope and security boundaries before merge.