Can I use this for non-test related tasks?

Yes, it applies to builds, documentation audits, and requirement checklists, requiring a line-by-line verification before any status update.

What counts as valid verification evidence?

Valid evidence requires fresh terminal output showing an exit code of 0, 100% test passes, or a direct confirmation of the fix; 'should work' or linter-only checks are insufficient.

Why is this skill necessary for AI development?

It prevents AI agents from making false success claims based on assumptions or previous states, ensuring every report is backed by real, current execution output.

How does this improve the developer experience?

It builds trust between the developer and the AI agent by guaranteeing that when the agent says a task is 'done,' it has actually been verified through code execution.

Verification Before Completion

Name: Verification Before Completion
Author: astrosteveo

byastrosteveo

0•

Security & Testing

Enforces a rigorous verification-first workflow that requires fresh command output before claiming any task is complete or successful.

Verification Before Completion acts as a critical gate function for Claude Code, preventing premature success claims and hallucinations regarding task status. By mandating fresh, empirical evidence for every claim—whether it's passing tests, fixing bugs, or meeting requirements—it ensures that the agent's reports are grounded in actual execution results rather than assumptions. This skill is essential for maintaining high software quality and high-trust automation by eliminating rationalizations like 'it should work' and replacing them with 'here is the passing output.'

Key Features

01Mandates fresh command execution before any success claim

020 GitHub stars

03Standardizes evidence requirements for builds, tests, and bug fixes

04Prevents premature completion reports in agentic development workflows

05Eliminates assumptions such as 'it should pass' or 'linter passed so tests passed'

06Implements a systematic Identify-Run-Read-Verify gate function

Use Cases

01Ensuring bug fixes are verified against original symptom tests before closing a task

02Guaranteeing that 'all tests pass' claims are backed by 100% current success rates

03Validating that deployment-ready code builds successfully without relying on partial checks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add astrosteveo/harness verification-before-completion

For use in Claude.ai and ChatGPT

Download Skill