diagnoselisted

# Diagnose A discipline for hard bugs and regressions. Skip phases only when explicitly justified. All questions and findings reported to the user must be in **Portuguese (pt-BR)**. When exploring the codebase, use the project's domain glossary from `.claude/CONTEXT.md` (if it exists) to get a clear mental model of the relevant modules, and check `docs/architecture/` for decisions and ADRs in the area you are touching. **Re-validation loop**: after every hypothesis, fix, or change, re-run the reproduction and the regression checks before declaring the bug resolved. **Iron rule:** NO FIXES WITHOUT A RED LOOP AND A CONFIRMED ROOT CAUSE. A change made before Phase 1 produced a red signal and Phase 4 confirmed a hypothesis is a guess — and guessing under pressure is how one bug becomes three. This holds ESPECIALLY under time pressure, when the bug "looks simple", and after a previous fix already failed: those are exactly the moments the process pays off. Violating the letter of this process is violating the spirit of debugging. ## Pipeline ```text Phase 0 Frame the bug → symptom, expected vs actual, env, frequency, scope Phase 1 Build a feedback loop → ONE tight, red-capable command (the core of this skill) Phase 2 Reproduce + minimise → smallest scenario that still goes red Phase 3 Hypothesise → 3–5 ranked, falsifiable predictions Phase 4 Instrument → one probe per prediction, one variable at a time Phase 5 Fix + regression test →