I built this because single AI agents often make inconsistent or risky decisions on code changes, especially in PR reviews where one hallucination or missed detail can ship bugs or vulnerabilities. The idea is simple: do not trust one model. Run multiple specialized agents in parallel and resolve the result through weighted consensus, with reputation scores improving over time. You can also add risk thresholds and human approval when needed. It acts like a decision firewall for agent actions, starting with GitHub PR merge guards. Runs fully local with SQLite and localhost. No cloud required. Would love feedback. Have you seen reliability issues with agents in your workflows? Try the PR guard template and tell me what breaks or what should be added next. GitHub issues and PRs welcome https://github.com/kaicianflone/...

Consensus MCP Board

Decision firewall for AI code reviews and PR merges

Decision firewall for AI code reviews and PR merges