What problem does it solve? Code changes and agent outputs can silently introduce regressions or low-quality work that only surfaces later in production. This Skill provides a verification layer that scores output reliability, runs regression guards in CI, and rolls back changes that fail a configurable quality threshold. ## Core Features & Use Cases - Truth Scoring: Computes 0.0-1.0 reliability metrics for code, agents, and tasks, with trends, confidence intervals, and table/JSON/CSV/HTML export formats. - CI Regression Guards: Ships three concrete layers — behavioral smoke tests (install, hooks, MCP protocol, memory import), a monotone-decreasing tool-description audit baseline, and Ed25519-signed witness manifests verifying documented fixes remain in dist builds. - Automatic Rollback: Reverts changes that fail verification against a configurable threshold (default 0.95), with selective, dry-run, and backup-first modes. - Use Case: A team adds a new MCP tool and pushes; the tool-descriptions audit fails CI because the description lacks usage guidance, and the witness verifier confirms all previously documented fixes are still present before publish. ## Quick Start Run a verification check on my current project and show me the truth scores for the last 7 days.