Pullfrog

Key features

• Status-check pull-request reviews Every incoming pull request gets an automatic structured review, and the verdict can gate merges as a requireable check, so review becomes a merge requirement rather than a comment thread.[1]

• CI and merge-conflict autofix Pullfrog detects failures on its own pull requests and attempts fixes, extends the same repair to human pull requests when configured, and resolves merge conflicts automatically to keep branches mergeable.[1]

• Issue triage and plan drafting New issues get responses to common questions, labels, links to related issues and pull requests, or drafted implementation plans, moving intake work off the maintainers' queue.[1]

• Ad-hoc prompts from anywhere Tagging @pullfrog in any issue, pull request, or comment pulls in the surrounding context for one-off work, and the hosted console accepts prompts for anything else, with each automation toggleable per trigger.[1]

• Real-agent wrapping, not a clone Runs use the vendor's actual Claude Code, Codex, or OpenCode agent with your existing repo configuration, so agent improvements and your repo's own skills flow through without re-teaching a separate bot.[1]

• Flexible model billing Connect an existing Claude, Codex, Grok, Kimi, or OpenCode subscription with no API key, bring any provider key for Anthropic, OpenAI, Google, xAI, Mistral, DeepSeek, or OpenRouter, or route through Pullfrog's gateway at raw provider cost with no markup.[1][2]

• GitHub-native security posture A purpose-built MCP server handles git and GitHub operations, every run uses a short-lived installation token that auto-revokes on completion, keys are masked in logs, only minimum-necessary environment variables pass through, and GitHub stays the single source of truth for access control.[1]

• Headless browser for UI work End-to-end test runs, screenshots with uploads, and UI iteration work out of the box, so front-end changes get exercised rather than only described.[1]

Who will love it

• GitHub teams buried in pull-request queues Automatic review on every incoming pull request with merge-gating status checks gives authors feedback before a human opens the diff, inside the workflow they already use.[1]

• Developers already paying for agent subscriptions Claude Pro/Max, ChatGPT Codex, and similar plans connect directly with no per-seat reviewer bill, so existing spend covers the automation.[1]

• Open-source maintainers Personal accounts and open-source repositories run free, with a path to apply for covered model costs, matching maintainer budgets of zero.[1]

Who should avoid it

• Teams outside GitHub The GitHub App plus Actions workflow is the only documented runtime, so GitLab, Bitbucket, and other forges have no supported path.[1][2]

• Organizations wanting a single vendor bill Model spend stays with the team's own provider or subscription on top of the Pro plan, so finance teams that require one invoice per tool will find the split billing awkward.[1]

• Teams wanting zero-configuration review Someone still installs the App, commits the workflow, chooses billing, and tunes per-trigger instructions, so there is real setup and upkeep.[1][2]

Alternatives and compatibility