Skip to content

AIgency verification pipeline

Verification captures real command evidence, evaluates it against acceptance criteria, records defects, and drives bounded repair/reverify cycles.

AgentDock Pro

AIgency is available only in AgentDock Pro.

aigency-verification-service.ts is what keeps AIgency from trusting an agent's own "I'm done" message. It's a pipeline, not a single check.

The pipeline

  1. Command evidence — a verification-kind task runs the configured check and captures the real output as evidence.
  2. Criterion evaluation — each acceptance criterion is scored against that evidence: pending, running, passed, failed, blocked, or skipped.
  3. Verification reports — aggregate the per-criterion results for a task or run.
  4. Defects — a failed required criterion is recorded as a defect with its own status lifecycle (open → repairing → resolved / accepted / rejected / cannot_reproduce).
  5. Repair iterations — a defect can drive a bounded repairing → reverifying cycle before landing on resolved, limit_reached, or stopped.
  6. Completion recommendation — the pipeline's output feeds the quality gate decision on whether a task or run is actually acceptable.

Related