Pullfrog

Pullfrog listens for GitHub events, pull requests opened, issues created, reviews submitted, CI failures, and triggers agent runs from a workflow file in your repository. Because the runs happen in your Actions, you keep the infrastructure, the keys and the bill, which is the part that usually decides whether a review bot is worth adopting.

It is not an agent itself. It wraps Claude Code, Codex and OpenCode, picking whichever matches your configuration, so every run uses the vendor's real agent and reads the repo-level setup you already keep for it: CLAUDE.md or AGENTS.md, skills, custom commands and MCP servers.

Each automation toggles from the dashboard and takes its own per-trigger instructions.

• Review new PRs: incoming pull requests get an automatic review whose verdict can gate merges as a required status check.
• Address review comments: leave feedback as you would for a colleague and it works through the thread.
• Autofix: failing CI on its own pull requests and merge conflicts are fixed without hand-resolution.
• Triage issues: apply labels, answer common questions, link related work or draft an implementation plan.
• Ad hoc requests: tag it in any issue, pull request or comment and it pulls in the surrounding context.
• Contained execution: GitHub operations use a short-lived installation token that is revoked when the run ends, shell commands run in an isolated subprocess, and keys are masked in logs.

Setup takes about two minutes through one npx command or the browser console. It is free for personal accounts and open source repos.