| name | deep-research |
| description | Deep domain research and knowledge synthesis — autonomously ingest, synthesize, and structure new technical domains |
| license | MIT |
| compatibility | opencode |
| metadata | {"audience":"researchers","workflow":"research"} |
Deep Research Skill
Description
A methodology for The Documentarian or The Research Agent to autonomously ingest, synthesize, and structure new technical domains into the local knowledge base.
Triggers
- When the Scribe detects a compilation error related to a missing or misunderstood library.
- When the user explicitly commands
/research [Topic].
- When an unknown protocol is introduced to the
opencode workspace.
Workflow
- Query Generation: Formulate specific, highly-targeted technical queries based on the gap in understanding.
- Ingestion (No-API):
- Use the
fetch MCP tool to retrieve raw documentation from official sources.
- If interactive pages are needed, utilize the
puppeteer MCP to render and scrape the DOM.
- Synthesization:
- DO NOT dump raw code.
- Summarize the findings into Conceptual Algorithms and Decision Matrices.
- Note the trade-offs (e.g., "WebSockets vs Systems IPC Events for high-frequency hardware events").
- Knowledge Crystallization: Write the findings into
C:\opencode\.opencode\skills\[domain]\references\. If it's a new domain, create the skill directory. Use the template provided in assets/tech-brief-template.md to ensure standardized formatting.
Assets
assets/tech-brief-template.md: The required structure for generating a Technical Brief after research concludes.
Best Practices
- Focus on constraints, limitations, and edge cases. (e.g., "What happens to the UI store if the Systems backend drops the event stream?").
- Always cite the URLs ingested via
fetch so The Architect can review the source of truth if necessary.