fast-review

What problem does it solve?

Fast Review prevents unchecked agent outputs from progressing in automated workflows by providing a lightweight, mandatory verification step that catches regressions, boundary violations, and semantic errors before acceptance. It reduces risk during overnight or long-running autonomous sessions by ensuring each delegation is validated quickly and consistently.

Core Features & Use Cases

• Single-reviewer sub-agent that evaluates outputs against acceptance criteria and ignores the original delegation prompt.
• Structured verdicts: outputs must follow a PASS | FAIL format with ISSUES, FEEDBACK, and CONFIDENCE fields.
• Automatic retry and escalation: retry the same agent up to two times on FAIL, escalate to a 3-reviewer panel on repeated failures, and record disputes for human resolution.
• Integration with deterministic gates: runs after lint/test/build, secret scanning, and blast-radius checks, and requires logging before proceeding.
• Use cases: rapid validation for routine code changes, overnight autonomous runs, and gating low‑ to medium‑risk delegations while escalating high‑stakes changes.

Quick Start

Run fast-review on the agent's output to obtain a PASS or FAIL verdict with structured issues, feedback, and recommendations for retry or escalation.