debugging-feedback-looplisted

# Debugging — build the feedback loop first ## Overview `superpowers:systematic-debugging` tells you to reproduce the bug consistently and form a hypothesis before fixing. This skill is the missing **HOW**: constructing the reproduction (feedback) loop that makes hypothesis-testing cheap. **The loop is the skill — everything after it is mechanical.** Do not proceed to hypothesise without one. Adapted from mattpocock/skills `engineering/diagnose`. ## When to Use - You have a bug but can't reliably reproduce it, or each reproduction is slow/manual. - The bug is non-deterministic / flaky / timing-dependent. - You're about to start guessing because reproduction is painful — stop and build a loop. Do NOT use this as a *replacement* for systematic-debugging — run that for the overall discipline (root cause before fix). Use this for the reproduce step inside it. ## The feedback-loop ladder (build the highest rung you can) Pick the fastest, most deterministic loop available: 1. A failing automated test at the bug's level (unit / integration). 2. A single `curl` / CLI command that triggers it. 3. A CLI snapshot diff (capture output, compare to expected). 4. A headless-browser script (Playwright) for UI bugs. 5. A trace / log replay against a captured request. 6. A throwaway harness script that drives just the failing path. 7. A property / fuzz test that searches inputs for the trigger. 8. A `git bisect run` harness (a script that exits 0/1) to find the breaking commit. 9. A differ