# Investigate

> Run (or re-run) Stage 1 only — the Problem Investigator: citation graph, elite pool, deep reading, and the grounded Research Brief. Operates on the active run, or scaffolds a new one if a topic is given.

- Skill: `emmanuelrtm/investigate` (Agent Skill)
- Install (CLI): `npx skillmds@latest add emmanuelrtm/investigate`
- Raw SKILL.md: https://api.skillmd.com/api/skills/emmanuelrtm/investigate/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: EmmanuelRTM (https://skillmd.com/u/emmanuelrtm)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/emmanuelrtm/investigate

---


# /investigate — Stage 1: Problem Investigator

Arguments: `$ARGUMENTS`

Current state: !`python3 .claude/scripts/ledger.py status`

1. Resolve the run:
   - `--run <id>` given → use `workspace/runs/<id>`.
   - A topic string given → scaffold a fresh run first:
     `python3 .claude/scripts/new_run.py $ARGUMENTS`.
   - Neither → use the ACTIVE_RUN shown above; if none, ask for a topic.
2. Launch the `problem-investigator` subagent with the run directory, topic,
   and offline flag from `run-config.json`. It follows its own three-round
   protocol and the relevance gate.
3. On `ok: false` (gate abort): relay the reason; suggest better seeds in
   `workspace/seeds/` or a broader topic.
4. On success: run `python3 .claude/scripts/bib_validate.py <run>` (must
   pass), append ledger event
   `{"event":"stage_investigate","detail":"complete"}`, and show the user the
   brief's section headers plus the source count.

Next stage: `/discover`.