hunt-defectlisted

# Hunt a Defect to Its Root Cause The default failure mode in debugging is pattern-matching the error message to a plausible-looking line, changing it, seeing the symptom disappear, and calling it fixed. That produces a codebase where the same bug returns in a different shape three weeks later. This workflow makes that impossible by forcing the order: **reproduce, then hypothesise, then test the hypothesis, then fix.** The hypothesis is stated and written down before any code is touched, so it can be wrong out loud. ## Step 0 — Understand the report - If `$ARGUMENTS` is empty, ask for the bug: what was expected, what happened, and how to trigger it. Stop until they answer. - If it looks like a file path, read it as a log or trace. - Otherwise take it verbatim. Extract and state back, in one short block: **expected behaviour**, **observed behaviour**, **trigger**, and the **error text with its stack trace** if there is one. Anything you cannot extract, name as unknown rather than filling in. Then read the code named in the trace, from the innermost frame that belongs to this project outward. Not just the line — the whole function, and its callers. ## Step 1 — Reproduce it before anything else **No hypothesis, no diagnosis, no fix until the bug is reproduced or reproduction is confirmed impossible.** A fix for a bug you have not seen fail is a guess. 1. Determine the smallest command that should trigger it. Prefer an existing test, then a single test case, then a scrip