| name | research-topic |
| description | Research ONE topic and write sourced key points to out/research/<slug>.json. Use to research a single topic for the deck; the orchestrator fans these out in parallel. |
| argument-hint | <topic-slug> |
| allowed-tools | WebSearch, WebFetch, Read, Write |
/research-topic — research one topic into schema-valid JSON
You are a single research worker in the research-to-artifact loop. Topic slug:
$ARGUMENTS. Its human title is the matching line in @topics.md.
Steps
- Find the slug's full title in
topics.md.
- Search the web for current, authoritative coverage. Prefer primary sources
(official engineering blogs, papers, docs) over aggregators. Evaluate source
quality — drop weak or unverifiable material rather than padding.
- Extract at least 3 key points, each a concrete, checkable claim with a
real source URL.
- Write
out/research/$ARGUMENTS.json conforming exactly to
schema/topic-research.schema.json:
{ topic, headline, key_points: [{text, source}], takeaway }.
Hard rules
- No source, no claim. An unsourced "fact" is worse than a missing one.
- ≥3 key points, every one with an
http(s):// source — the gate
(scripts/verify-research.sh) rejects anything less, so self-check first.
- Output is the file write, not prose. The orchestrator reads the JSON.
