differential-debugging-and-bisectionlisted

# Differential debugging and bisection `superpowers:systematic-debugging` covers the general hypothesis loop for a bug you can already see and reproduce. This skill covers the gaps: bugs defined by a *change over time* (regressions), a *change over space* (works-here-fails-there), or a *change over runs* (nondeterminism). Load this alongside systematic-debugging, not instead of it — use it to locate the cause, then hand off to that skill's hypothesis loop to fix it. ## 1. Regression → git bisect first If the bug is "this used to work," don't hypothesize from the current diff — bisect finds the actual causal commit in O(log n). 1. Confirm you have a known-good ref and a known-bad ref (a tag, an old commit, `HEAD`). 2. `git bisect start`, `git bisect bad <bad-ref>`, `git bisect good <good-ref>`. 3. If the repro is scriptable (exit 0 = good, non-zero = bad, 125 = skip this commit — e.g. it doesn't build), automate it: `git bisect run <script>`. Writing that script IS the minimal-repro step (§3) — do it before bisecting, not after. 4. `git bisect view` / the final output names the first-bad commit. `git bisect log` captures the trail for the eventual root-cause writeup. Always `git bisect reset` when done, even on failure paths. 5. The first-bad commit is a strong causal hypothesis, not a proven root cause — read its diff, then verify with systematic-debugging's evidence loop rather than assuming the commit message explains the bug. ## 2. Differential