verification-before-completion

Enforces running fresh verification commands before claiming work is complete.

1|Updated Jul 18, 2025
One-click install
npx skills add https://github.com/gusgad/service-agent-by-sap --skill verification-before-completion-gusgad
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: verification-before-completion
Source: https://github.com/gusgad/service-agent-by-sap/tree/main/.agents/skills/superpowers/verification-before-completion
Command: npx skills add https://github.com/gusgad/service-agent-by-sap --skill verification-before-completion-gusgad

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI agents and developers often claim work is done, tests pass, or bugs are fixed based on stale memory or assumptions rather than fresh evidence. This Skill enforces a strict gate: no completion claims without running the actual verification command and reading its output in the current turn. ## Core Features & Use Cases - Evidence Gate: Requires identifying, running, and reading the output of the exact command that proves a claim (tests, lint, build, bug reproduction) before stating success. - Rationalization Detection: Lists common excuses like "should work now" or "linter passed" and maps them to why they are not evidence. - Regression Test Discipline: Defines a red-green pattern where a regression test must be observed failing without the fix and passing with it. - Use Case: Before committing a bug fix, the agent re-runs the exact failing scenario, confirms the symptom is gone, and only then reports the fix as verified. ## Quick Start Before claiming the task is done, run the relevant test or build command fresh and report the actual output as evidence.

Frequently Asked Questions about verification-before-completion

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify code changes before committing?▼

Identify the command that proves your claim, such as npm test or npm run build, run it fresh in the current turn, and read the full output including exit code and failure counts. Only state success after the output supports the claim.

What counts as evidence that a bug is fixed?▼

A bug is fixed only when the original reported symptom is re-tested and confirmed gone. Changing the code and assuming it works is not evidence; you must reproduce the exact scenario that failed before.

Does a passing linter mean the build succeeds?▼

No, a linter passing is not evidence the build compiles. Linters check style and some errors but do not perform compilation, so you must run the actual build command and check its exit code.

How do I write a proper regression test?▼

Write the test, confirm it passes with the fix, temporarily revert the fix to confirm the test fails, then restore the fix and confirm it passes again. A regression test never observed failing has not proven anything.

Can I trust a subagent's report that a task succeeded?▼

No, delegated work requires independent verification. Inspect the actual VCS diff or command output yourself before repeating a subagent's success claim to the user.