How does the verify skill work within a workflow?

It is an internal runner called by the /do command that parses a project manifest and spawns specialized agents to verify all criteria in parallel.

Can I run /verify directly as a user?

No, the verify skill is marked as non-user-invocable and is designed to be triggered by orchestration commands to ensure a consistent verification process.

How does it handle manual verification steps?

If all automated checks pass but manual criteria exist, the skill lists the manual requirements and suggests calling /escalate for human review.

What happens if a verification check fails?

The skill provides a structured report including the specific failure location, the difference between expected and actual results, and actionable hints to resolve the issue.

What are Global Invariants in this context?

Global Invariants are critical system-wide rules that must pass for any task to succeed; if these fail, the skill highlights them as primary blockers.

Automated Manifest Verification

Name: Automated Manifest Verification
Author: doodledood

bydoodledood

•

安全与测试

Orchestrates parallel agentic verification of project requirements, global invariants, and acceptance criteria.

The verify skill acts as an automated quality assurance engine for Claude Code workflows by parsing project manifests and spawning specialized sub-agents to validate technical requirements in parallel. It categorizes checks into shell commands, codebase patterns, and reviewer agents, ensuring that global invariants—system-wide constraints that must always hold true—and specific acceptance criteria are met before a task is considered complete. This skill streamlines the development lifecycle by providing actionable feedback on failures and facilitating human escalation when manual verification is required, significantly reducing the overhead of manual code review.

主要功能

018 GitHub stars

02Parallel execution of multiple verification criteria to minimize latency

03Prioritized reporting of Global Invariant failures to block invalid releases

04Support for bash commands, sub-agent reviews, and codebase pattern matching

05Intelligent routing for manual verification steps requiring human oversight

06Detailed failure feedback including file locations, expected values, and fix hints

使用场景

01Orchestrating multi-agent QA workflows to verify acceptance criteria simultaneously

02Validating complex software changes against a predefined project manifest

03Enforcing architectural invariants and coding standards during automated refactoring

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add doodledood/claude-code-plugins verify

For use in Claude.ai and ChatGPT

Download Skill