| name | diagnosing-bugs |
| description | Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose" / "debug this", or reports something broken, throwing, failing, or slow. |
diagnosing-bugs
A discipline for hard bugs. Build a tight loop that goes red on this
bug FIRST; everything after is mechanical. Doctrine + the fleet adaptations:
diagnosing-bugs.md. Skip a
phase only with an explicit reason.
Phase 1 — Build the loop
Construct one named, agent-runnable command that drives the bug code path and
asserts the user's exact symptom — a failing pnpm test <file>, a CLI+snapshot
diff, an httpJson script, a trace replay, a bisection/differential loop. Never a
throwaway harness you delete after.
Make it tight: fast (seconds), deterministic (pin time, seed RNG, freeze the
network with nock), sharp (asserts the symptom, not "didn't crash"). For a flaky
bug, raise the reproduction rate until it's debuggable.
Completion criterion: you can paste one command you have already run and
its output, and it is red-capable, deterministic, fast, and agent-runnable. If you
catch yourself reading code to build a theory before this command exists — stop.
No red-capable command, no Phase 3.
If you genuinely cannot build a loop, say so explicitly, list what you tried, and
ask for environment access / a captured artifact. Do not proceed to hypothesise.
Phase 2 — Reproduce + minimise
Run the loop; watch it go red on the user's symptom (not a nearby one). Then
shrink the repro: cut inputs, callers, config one at a time, re-running after each,
until every remaining element is load-bearing. Don't proceed until reproduced AND
minimised.
Phase 3 — Hypothesise
Generate 3–5 ranked falsifiable hypotheses before testing any — each states a
prediction ("if X is the cause, changing Y makes it disappear"). No prediction =
a vibe; sharpen or discard. Show the ranked list to the user before testing
(cheap checkpoint; they may re-rank instantly) — but proceed on your ranking if
they're AFK.
Phase 4 — Instrument
One variable per probe, mapped to a specific Phase-3 prediction. Prefer a
debugger/REPL over logs. Tag every debug log [DEBUG-<hex>] so cleanup is one
grep — never "log everything and grep". Perf bugs: measure a baseline first
(performance.now(), profiler, query plan), then bisect.
Phase 5 — Fix + regression test
Write the test before the fix, but only at a correct seam — one that
exercises the real bug pattern at the call site. Fleet seam doctrine: public
interface, highest available seam, vitest in test/ (members co-locate; wheelhouse
hook/lint-rule tests in test/repo/), never node:test, never source-text
assertions. Turn the minimised repro into a failing test, watch it fail, apply the
fix, watch it pass. If no correct seam exists, that is the finding — flag it.
Phase 6 — Cleanup + post-mortem
grep out the [DEBUG- tags, run pnpm run check + pnpm test, commit with the
root cause named in the message (fix(<scope>): <root cause>). If this is the
second occurrence of the same shape, search the repo for siblings before closing
(variant analysis).
