Coverage proves a line ran; it does not prove a test would catch the line breaking. This loop attacks test QUALITY rather than quantity: each pass injects one small semantic mutation into recently changed code (flip a boolean, drop a guard, change an operator, weaken a threshold) and runs the suite — any mutant that survives is a test encoding behavior but not intent, so the loop strengthens the missing assertion, reverts the mutant, and repeats. Distinct from autoloop-tdd (writes new behavior test-first) and coverage-until-threshold (hits a percentage): here the exit condition is that every injected mutant is killed.
Loop
- Goal
- every semantic mutation injected into recently changed code is caught by a failing test (no surviving mutants), with the suite green and all mutants reverted
- Exit when
- every manually injected mutant produced a failing test, the suite is green, and all mutants are reverted (a clean run confirms no mutant is left behind)
Check between iterations
npm test
$
Kickoff prompt
Start the "Intent-Test Mutation Guard" loop.
Goal: every semantic mutation injected into recently changed code is caught by a failing test (no surviving mutants), with the suite green and all mutants reverted
Max iterations: 10
Between iterations run: npm test
Exit when: every manually injected mutant produced a failing test, the suite is green, and all mutants are reverted
Step 1: Pick one recently changed function, inject a single semantic mutation (flip a condition, drop a guard, or weaken a threshold), run the suite, and if it stays green strengthen the test that should have failed before reverting the mutant.
›