research

/research — Background Research via Argus

Research is dispatched to workers. Claude coordinates, not executes.

CRITICAL: Dispatch Is Mandatory

Research queries MUST go through Argus or Gemini CLI as a subprocess. Claude does NOT do web research inline. Build the query, dispatch, integrate results.

Usage

/research [topic]

Process

1. Ask 1-2 clarifying questions (goal, depth)
2. Check Argus availability:

   curl -s --connect-timeout 3 http://100.112.130.100:8270/api/health >/dev/null 2>&1 && echo "argus: yes" || echo "argus: no"
   

3. Dispatch research to worker (MANDATORY):
   • If Argus available, dispatch via the research lane:

     ~/github/oneshot/bin/dispatch \
       --class doc_draft \
       --category research \
       --prompt "Research: [topic]. Query Argus in research mode via HTTP API, extract key results, write findings to docs/research/[slug]/research.md. Structure: Executive Summary → Findings → Sources." \
       --output 1shot/dispatch \
       --manifest 1shot/dispatch
     

   • If Argus unavailable, dispatch to Gemini CLI:

     ~/github/oneshot/bin/dispatch \
       --class doc_draft \
       --category research \
       --prompt "Research: [topic]. Write findings to docs/research/[slug]/research.md." \
       --worker gemini_cli \
       --output 1shot/dispatch
     

4. Review worker output when dispatch completes
5. Integrate findings, present summary

Search Modes (from config/search.yaml)

• discovery: broad, multiple sources
• precision: targeted, high relevance
• research: comprehensive, all providers
• cheap: SearXNG only

Notes

• Research takes 3-8 minutes in background
• Check docs/research/ for past research
• Argus automatically handles provider fallback, ranking, and dedup