What is the 'Iron Law' of Shipyard Verification?

The Iron Law states: 'No completion claims without fresh verification evidence.' This means if a verification command hasn't been run in the current context, the work cannot be called done.

What happens if a verification command fails?

The skill mandates an honest report of the failures. The agent must state the actual status with evidence and address the failures rather than making a success claim.

How does Shipyard Verification prevent AI hallucinations?

It intercepts completion-related language and triggers a mandatory workflow that requires the agent to provide terminal output evidence before it is allowed to assert that a task is finished.

Can I use this with any test framework?

Yes, the skill is framework-agnostic. It instructs Claude to identify the specific commands relevant to your environment, such as npm test, pytest, or custom shell scripts.

Shipyard Verification Gate

Name: Shipyard Verification Gate
Author: lgbarn

bylgbarn

•

セキュリティとテスト

Enforces rigorous verification and evidence-based reporting before allowing tasks to be marked as complete, fixed, or ready for commit.

Shipyard Verification is a specialized discipline layer for Claude Code designed to eliminate premature success claims and 'hallucinated' passing states. By mandating a strict 'evidence before assertions' protocol, it requires AI agents to identify and execute specific verification commands—such as test suites, linters, or build scripts—and analyze the actual output before claiming a task is finished. This skill serves as a critical quality gate, ensuring that every code change is backed by fresh, verifiable results, thereby maintaining project integrity and developer trust.

主な機能

01Real-time detection of non-committal language like 'should' or 'probably'

02Mandatory execution of verification commands before any completion claims

03Comprehensive failure analysis to prevent reliance on partial or stale results

0438 GitHub stars

05Structured gate function for identifying, running, and reading command outputs

06Teammate Mode for reporting verification artifacts to lead agents via metadata

ユースケース

01Maintaining strict quality standards during parallel agent task execution

02Ensuring bug fixes are validated against original symptoms before PR creation

03Preventing broken or unverified code from being committed to a repository

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add lgbarn/shipyard shipyard-verification

For use in Claude.ai and ChatGPT

Download Skill