Problem
The proposal's core innovation is an iterative verification loop that repairs failing tests until pass. Reality: core/generator.py:285-322 runs tests once, records tc.diagnosis, and never applies a fix or re-runs. 'Self-healing' is currently a no-op during generation.
Tasks
Acceptance
A deliberately-wrong generated test gets diagnosed, auto-fixed, and passes on re-run within max iters; loop terminates and is unit-tested.
Problem
The proposal's core innovation is an iterative verification loop that repairs failing tests until pass. Reality:
core/generator.py:285-322runs tests once, recordstc.diagnosis, and never applies a fix or re-runs. 'Self-healing' is currently a no-op during generation.Tasks
max_repair_iters, default 3).TEST_WRONGdiagnosis: applydiagnosis.suggested_fix, regenerate test file, re-run.CODE_BUG: flag, stop repairing (do not mutate test to force pass).INTENT_WRONG: surface for intent revision (interactive) or skip (auto).VerificationResult.Acceptance
A deliberately-wrong generated test gets diagnosed, auto-fixed, and passes on re-run within max iters; loop terminates and is unit-tested.