Skip to content

desk-research

Evidence-grounded research that travels with you across every repo. Seven skills — scoping, source curation, synthesis, adversarial review, decision support, and rationale reconstruction — plus two read-only retrieval subagents. The methodology is grounded in seven convergent disciplines (STORM, PRISMA, ACH, Wikipedia V/RS/NPOV, OSINT, GIJN, GRADE).

What's inside

  • Skills for the full research loop: desk-research, source-map, build-outline, compare-hypotheses, identify-perspectives, devils-advocate, decision-archaeology.
  • evidence-retriever and source-extractor subagents (read-only).

Install

desk-research is user-scope by default — research method is portable, not project-specific.

agentbundle install --pack desk-research <catalogue>

It projects to every shipped adapter that supports the skill primitive (Claude Code, Codex, Copilot, Cursor, Gemini, Kiro).

Claude Code only — grant web tools to the retrieval subagents. On Claude Code, add WebSearch and WebFetch to your permissions.allow so the evidence-retriever and source-extractor subagents can do live web retrieval — a non-interactive subagent cannot surface the approval prompt, so without the grant those tools are denied. This is a one-time note, not bundle-managed machinery: permissions.allow is yours to own. Other adapters (Copilot, Cursor, Gemini, Codex, Kiro) pass the web tools through or bake them in at build time and need no such step.

Usage

Ask your agent, for example:

  • "Research how teams measure deployment frequency, and map the sources."
  • "Synthesise these five papers into a one-page briefing with citations."
  • "Run a devil's-advocate pass on the claim that X causes Y."

→ Go deeper: the desk-research guides.