Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →brenbuilds1/unclebob
A do-not-read-the-code workflow for delegating implementation to a coding agent.
Passed 21 of 23 test cases with the skill, and 9 of 23 without it.
and 18 more in the full report
12 fixed9 passed either way2 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
During our code verification round on a pull request modifying 200 lines of code, the overall repository test coverage is 95%, but the test coverage on the modified 200 lines is 82%. Should this pull request pass the coverage constraint?
| Checked | Without skill | With skill |
|---|---|---|
| Marks the coverage constraint as failed because changed-line coverage is below 90%. | Missed | Passed |
| Evaluates coverage specifically on changed lines rather than overall repository coverage. | Passed | Passed |
The instructions your model reads when the skill loads.
The promise of agent coding is that you stop reading the code. The problem with keeping that promise is that the constraints meant to replace your reading are written by the same agent that writes the code. A test suite authored by the thing it judges proves very little on its own.
This skill keeps the promise anyway, by making a faked pass cost more than an earned one. Scenarios get approved by the human before any code exists. Tests get written, shown to fail, and locked before implementation starts. After the lock, no round of work may touch code and tests together. And mutation testing runs at the end because it is the one constraint a vacuous test cannot talk its way through.
Skip it for throwaway scripts and exploratory spikes, and say so out loud when skipping it. The full gauntlet is for code nobody will read. When the human is going to read the diff anyway, plain unit tests can be the whole gauntlet; running every constraint on a two-line fix is ritual, not confidence.
test cases passed, vs 9 / 23 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.