deep-research

name	deep-research
description	Run deep research in a brainstorm-solutions workspace. Reads all context, executes queued prompts, and saves structured findings. Triggers on phrases like "deep research", "start researching", "run the research", "investigate this blocker".

Deep Research

Execute the research workflow inside a brainstorm-solutions workspace. This skill assumes you are already inside a research workspace created by the brainstorm skill.

Pre-flight

1. Confirm this is a research workspace — check for context/, prompts/, outputs/ directories and a research-brief.md.
2. Read all context files:
   • context/from-human/research-brief.md — the blocker and what was tried
   • Everything in context/from-history/ — prior compacted findings (if any)
   • Everything in context/from-internet/ — any saved reference material

Research execution

1. Check prompts/queue/ for the next prompt to run. If empty, check prompts/run/initial/ for unexecuted prompts.
2. Read the prompt file thoroughly.
3. Conduct research using all available tools:
   • Web search (Tavily, web fetch) for documentation, GitHub issues, forums, blog posts
   • Context7 for library/framework documentation
   • Reasoning to synthesize findings
4. Save output to outputs/individual/YYYY-MM-DD-<slug>.md with:
   • ## Key Findings section at the top
   • Detailed analysis in the body
   • ## Sources section at the bottom with URLs
5. Move the executed prompt from prompts/queue/ to prompts/run/initial/ (or subsequent/ if it builds on prior work).
6. If the findings suggest follow-up questions, create new prompt files in prompts/queue/ for subsequent investigation.

After research

1. Commit and push all new outputs and prompt movements.
2. Summarize findings to the user with:
   • Top 3 most promising approaches
   • Recommended next step
   • Any open questions that need further investigation
3. Ask if the user wants to:
   • Run another research iteration (deeper dive on a specific approach)
   • Compact findings into context/from-history/
   • Aggregate into a final report in outputs/final/