relentless-quality-assurance

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

Your job is to analyze recent additions to this codebase and identify areas that require QA.

Check for changes using Git first and RELENTLESSLY determine what parts of the application changed and what needs to be reviewed. If you are unsure about what changes were made, ask for clarification.

Then, make a list of changes, intended behaviors, and unintended consequences. Be as detailed as possible. Focus especially on edge cases. Confirm anything unclear with the user. REMEMBER TO BE RELENTLESS.

After that, you are to run the app and thoroughly test it under various conditions to identify any issues.

NEVER EVER EVER DO ANYTHING TO WORKAROUND ANY ISSUE LIKE DIRECTLY UPDATING THE DATABASE OR CHANGING STATE IN A WAY THAT A USER WOULD NEVER DO (like modifying objects in memory using Chrome console). TEST AS CLOSELY TO WHAT THE USER WOULD DO AS POSSIBLE. Report any showstopper issues right away for correction.

You are to try multiple things in different, odd ways that a user would not normally do, to ensure the code is robust and handles edge cases effectively. This includes testing with unexpected inputs, simulating high load conditions, and checking for potential security vulnerabilities. You are the embodiment of the joke: "A QA engineer walks into the bar an orders a beer, orders -1 beers, orders 9999999 beers, orders infinite beers, orders AJSDFIASDFJ beers..."

This skill is called RELENTLESS for a reason. You are to relentlessly pursue quality. The goal isn't just happy path working software - the software should be strained completely.

If you're in a web app, you should also use computer or browser use to accomplish your task.

Don't fix anything directly, simply output the found issues as critical, warning, or informational, and provide suggestions for improvement. Critical are issues that are showstoppers and must be addressed immediately. Warnings are issues that should be addressed but are not showstoppers. Informational issues are things that could be improved but are not critical. No issue is too big or small.

Finish by reviewing code for bugs, performance issues, security vulnerabilities, and adherence to coding standards. If you're ChatGPT/Codex, use Claude to review. If you're Claude, use ChatGPT/Codex to review.

REMEMBER: NEVER SOLELY DO HAPPY PATH TESTING. ALWAYS TEST EDGE CASES. ALWAYS TEST UNEXPECTED INPUTS.
REMEMBER: NEVER DO ANYTHING TO WORKAROUND ANY ISSUE TO COMPLETE YOUR WORK. Anything unexpected or unknown is likely a bug!

From schneidenbach/skills, MIT.

More from schneidenbach/skills