How does this skill prevent AI hallucinations in coding?

By enforcing an 'Iron Law' of fresh evidence, it requires the agent to see an exit code of 0 or zero failures before it is allowed to use words like 'Done' or 'Fixed'.

How does it handle tasks delegated to other agents?

It requires the primary agent to independently verify the VCS diffs and run local checks rather than blindly trusting success reports from sub-agents.

When should I use this Claude Code skill?

This skill should be active throughout the development process, specifically before every commit, pull request, or transition to a new task.

What is the primary purpose of the Strict Verification Gate skill?

It prevents Claude from claiming work is finished until it has successfully executed and verified the output of relevant commands like tests, builds, or linters.

Does this skill work with regression testing?

Yes, it specifically enforces Red-Green-Refactor patterns, ensuring a test fails when the fix is reverted and passes when it is restored.

Strict Verification Gate

Name: Strict Verification Gate
Author: EricGrill

byEricGrill

•

安全与测试

Enforces rigorous evidence-based verification before making any claims of task completion or success.

Strict Verification Gate is a mission-critical skill for Claude Code that implements a mandatory 'Gate Function' to prevent premature or false assertions of work completion. It requires the AI to identify, execute, and interpret fresh verification commands—such as unit tests, build scripts, or linters—before committing code or creating pull requests. By prioritizing objective evidence over assumptions, this skill eliminates common AI failures like shipping broken builds or missing edge cases, ensuring a high-trust relationship between the developer and the agent.

主要功能

01Red-Green-Refactor cycle enforcement for robust regression testing

022 GitHub stars

03Evidence-based reporting that links assertions directly to test and build outputs

04Strict requirement validation through line-by-line project checklists

05Mandatory gate function requiring fresh command execution before any success claim

06Independent verification of changes when delegating tasks to sub-agents

使用场景

01Preventing broken builds by verifying full compilation before PR submission

02Ensuring bug fixes are effective through verified TDD-style testing cycles

03Auditing project requirements to guarantee every feature is fully implemented

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ericgrill/agents-skills-plugins verification-before-completion

For use in Claude.ai and ChatGPT

Download Skill