Verdict

•
7
•

Provides an AI-powered QA agent that performs delta-based code quality assessments, identifying new and regressed issues against a historical baseline.

Verdict is a sophisticated Claude Code plugin designed to act as a skeptical QA agent, revolutionizing how AI-driven code quality assurance is conducted. Unlike typical AI agents that perform fresh, repetitive audits, Verdict maintains a historical baseline, enabling it to deliver delta-based reports that highlight new, regressed, resolved, or still-open findings. It offers granular defect classification, robust flaky test management with quarantines and expiries, and enforces a strict four-verdict release contract for CI. Emphasizing independence, Verdict operates read-only on your code with built-in security guards and provides detailed insights into coverage and residual risk, ensuring a thorough and defensible quality assessment without ever claiming 'no bugs found'.

主な機能

01Classifies failures into granular types like `REAL_DEFECT`, `FLAKY` (with quarantine & expiry), or `STALE_EXPECTATION`.
021 GitHub stars
03Enforces a four-verdict release contract (`pass`, `pass with risks`, `blocked`, `fail`) for CI/CD gates.
04Offers a read-only Model Context Protocol (MCP) server for external access to QA state and trends.
05Performs delta-based code quality assessment with historical baseline tracking.
06Operates read-only on the codebase, with hooks and Bash guards preventing unintended modifications.

ユースケース

01Automating continuous code quality assurance and release gating in CI/CD pipelines.
02Performing shift-left quality reviews on specifications and PRDs before code development.
03Intelligently managing and triaging flaky tests with automated quarantines and expiries.