What problem does it solve? AI-assisted delivery pipelines rely on prompt-level rules that an agent can talk itself out of. Guard enforces the rules a machine can settle — unchanged requirements, unweakened tests, append-only reviews, self-consistent verdicts, and real evidence — so violations are reported mechanically instead of being adjudicated away. ## Core Features & Use Cases - Mechanical rule checks: Verifies requirements immutability, test integrity (no deletions, skips, or only markers), append-only review records, verdict consistency, criterion coverage, evidence validity, and branch placement. - CI gate installation: Vendors the checker to .quorum/guard.py and writes a GitHub Actions workflow (with GitLab guidance on other hosts), plus a --check-gate mode that reports whether branch protection is actually live. - Self-testing checker: Ships a self-test suite that builds throwaway repositories, breaks each rule deliberately, and asserts the rule fires. - Use Case: Before merging an AI-produced pull request, run the guard to confirm no acceptance criterion was silently omitted from the verdict and no test was weakened, then make it a required status check so violations block merges. ## Quick Start Run the guard checks on my current work item and report any violations found.
