/debug

$ARGUMENTS

No code writing in this skill — only investigation and a fix plan. Find the real root cause in this codebase, not a plausible-sounding guess.

Steps

1. Update code graph: Run buddhi generate via the terminal-runner agent to ensure the code graph (.buddhi/graphs/tree-graph.db and tree-graph.json) is up to date before investigation.
2. Gather the symptom from $ARGUMENTS. If any of the following is missing, ask one question at a time — the error/symptom, reproduction steps, expected vs. actual behavior, and what changed recently (check via git log, run through the terminal-runner agent) — stopping and waiting for the answer before asking the next. Do not batch questions.
3. Before reading raw source, consult .buddhi/docs/ and use Buddhi MCP tools (buddhi_search, buddhi_read, see the okf-context skill) or query .buddhi/graphs/tree-graph.db / tree-graph.json for the subsystem the symptom points to, per the harness's docs/graph-before-source convention.
4. Form a ranked list of candidate root causes, most likely first, based on what step 3 turned up. If the failing area spans domains ambiguously enough that a /plan-style breakdown would help, optionally dispatch the relevant specialist agent(s) under agents/ to narrow the candidates — this is a judgment call, not a required step.
5. Investigate systematically rather than jumping at the first plausible hypothesis: reproduce the failure via terminal-runner where possible, inspect real logs/output, and eliminate candidates one at a time using that evidence.
6. Once the root cause is confirmed, report it with file:line references drawn from the graph/docs (not vague descriptions), plus a concrete fix plan — what needs to change and where. Do not write or apply the fix. Suggest running /verify once the fix has actually been applied, to prove it worked.