| name | deep-research |
| description | Run a thorough, source-heavy investigation on any topic. Produces a cited research brief with provenance tracking. |
| version | 1.0.0 |
Deep Research Skill
Run a comprehensive research investigation with evidence gathering, drafting, citation, and review.
Workflow
- Plan - Derive a slug from topic, create plan artifact
- Gather Evidence - Search papers, web, repos using researcher agents
- Draft - Write report from research notes
- Cite - Add inline citations, verify URLs
- Review - Simulated peer review with severity grading
- Deliver - Final output with provenance sidecar
Output Convention
All files use the slug prefix:
- Plan:
outputs/.plans/<slug>.md
- Draft:
outputs/.drafts/<slug>-draft.md
- Cited:
outputs/.drafts/<slug>-cited.md
- Output:
outputs/<slug>.md
- Provenance:
outputs/<slug>.provenance.md
Provenance Requirements
Every deep research output must include a .provenance.md sidecar with:
- Evidence table with numbered sources
- Verification status (PASS / PASS_WITH_NOTES / BLOCKED)
- Sources consulted, accepted, rejected
- Open questions and blocked checks
Scale Decision
- Direct search: narrow "what is X" questions, 3-10 tool calls
- Subagents: broad surveys, multi-faceted topics, 3-6 researcher agents
Integrity Rules
- Never fabricate a source
- URL or it didn't happen
- Mark verification status honestly
- Remove unsupported claims, don't smooth over gaps