Mullet
experimental - business in the front party in the back
Create tests that repay their maintenance.
Verify every change. Keep lasting coverage only when it protects a real, durable risk.
npx skills add ziptied/mullet --skill mulletProvider-aware proof for Pest Agent/Evals. Cursor audits. Same Mullet ladder.

## the_decision
LLMs test to prove the prompt. Your repo pays to keep the proof.
Agents love to add tests, introduce frameworks, and invent new ways to verify the change they were asked to make.
The problem is not verification. The problem is that those tests keep costing time, tokens, CI minutes, and refactor effort long after the prompt is done. Mullet makes every permanent test prove it protects a real, durable risk. Otherwise, verify that the work works and stop there.
Name the risk
What important bug would this test prevent later?
Price the cost
Will it repay future runs, repairs, reviews, and tokens?
Stop at proof
If the risk is not durable, validate the change and move on.
Verify everything. Keep only durable protection.
## the_test_debt_flywheel
Temporary proof turns into permanent drag.
The test may be useful for the current task. That does not mean it deserves to live in the suite.
Prompt
The agent adds a test to prove the task is complete.
Suite
That proof becomes part of the permanent project.
Cost
Every future developer, run, and agent has to carry it.
admission_gate
Mullet blocks test debt before it becomes infrastructure.
Verify every change. Keep only tests grounded in durable risk.
## evidence
Protect first. Cut second.
683 Pest files reviewed under Mullet full. The serious gaps come first; deletions wait for confirmation.
- 01
Add the 3 serious-path sentinels
- 02
Remove low-signal source-shape tests
- 03
Fold duplicate KYC + enrollment checks
Kept coverage: money, holds, refunds, fraud, enroll gates, OAuth, identity upsert.
then confirm removals
## intensity
Pick how hard Mullet pushes back.
Most agents treat every prompt as a reason to add coverage. Mullet changes the default: verify the work, then keep only tests that protect a risk that will still matter later.
Mullet lite
rigorous
A denser safety net for mature or compliance-sensitive products.
Mullet full
balanced · default
Positive evidence that risk and durability repay the test.
Mullet ultra
canary
Only the smallest faithful sentinels for serious, durable risk.
One line. Every supported agent.
npx skills add ziptied/mullet --skill mulletSame rule: verify every change, keep only tests that repay their maintenance.


