Find cause

a. Locate. Until the cause is proven this skill outranks spec and build; a read-only planning turn writes the plan per ../spec/references/task-list.md, reproduction test as Task 1. Worked before and one command reproduces it: run the bisect in investigator-prompt.md, then run node "${CLAUDE_SKILL_DIR}/../../lib/scratch-exclude.mjs" (with no symbol, exo:research-worker with Sources: repo). b. Investigate. Proven needs a causal line plus a mechanism predicting it or holding in source: write Status: proven. Otherwise dispatch exo:build-task with exo-kind: hardest from investigator-prompt.md. c. Fix. Settle where it commits per ../build/references/workspace.md, else copy the handoff to .exo/debug/. Dispatch general-purpose from fixer-prompt.md, the prompt opening with exo-kind: build and a Reads: line, resume failed via SendMessage, else dispatch a fresh agent with the prior result path. d. Report. Read only status lines and Report fields. After compaction, re-run Repro.

1. Reproduce. Read an existing log, profile or .exo/ report first. One command reproduces it, else evidence, no fix.
   • Send long output to <scratch>/debug-repro.log, <scratch> from node "${CLAUDE_SKILL_DIR}/../../lib/scratch-path.mjs" debug; read with tail -n 40.
   • Steps 1-5 → read references/loop.md first.
2. Instrument.
3. Isolate.
4. Predict, then fix.
5. Prove.
6. Retain project knowledge. Per the table.
7. Fresh eyes. With no PR review, read Step 7 of references/loop.md, which sets code-review for when it prints changed files above 2 or dependency-added yes, or the fix crosses a public signature, persisted format or security boundary, then ends on ship.
   • Invoked by a stage or workflow: ask nothing, return to caller.

References

Report: the report rule in ../../lib/references/closing.md with the debug log path as the pointer; mechanism, proof lines and SHAs go to the log; unproven: Repro, Expected, Actual, Hypotheses.