Skip to content
agent-ready-repo
GitHub

Desk Research

Your agent can research a question from sources it cites, grade how sure it is, and keep a long investigation going across sessions.

Install

Section titled “Install”

You need agentbundle first. Install it once for your user account:

agentbundle install --pack desk-research

In Claude Code you can install it as a plugin instead:

claude plugin install desk-research@agent-ready-repo

What your agent can do

Section titled “What your agent can do”
  • Answer a look-up, fact-check, or survey with citations, at a depth you pick from quick to deep.
  • Map the authoritative sources and the competing viewpoints on a topic before it searches.
  • Compare competing hypotheses in an evidence matrix.
  • Argue against its own findings and downgrade claims that do not hold up.
  • Run a research project over days or weeks, then write a brief with each claim graded by evidence quality.
  • Reconstruct why a past decision was made from commits, pull requests, and design docs.

Skills

Section titled “Skills”
  • desk-research — Researches one question in one session, with citations and graded claims.
  • source-map — Curates the authoritative sources on a topic before research starts.
  • build-outline — Breaks a question into the sub-questions a full answer must cover.
  • identify-perspectives — Names the camps on a contested topic before research begins.
  • compare-hypotheses — Scores competing hypotheses in an evidence matrix.
  • devils-advocate — Searches for counter-evidence and names the strongest objections.
  • decision-archaeology — Reconstructs the reasoning behind a past decision.
  • desk-research-project-start — Starts a multi-week research project.
  • desk-research-project-status — Shows the project’s phase, working hypothesis, and next step.
  • desk-research-project-check — Reports whether new sources have stopped changing your picture.
  • desk-research-project-digest — Builds the synthesis matrix and notes from your sources.
  • desk-research-project-synthesize — Writes the verdict and a self-contained brief.

Subagents, called for you: evidence-retriever fetches sources with citations, and source-extractor pulls the content out of each one. Both are read-only.

Works with

Section titled “Works with”

Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and Kiro IDE. Product Strategy reads your survey as evidence for its stakeholder research. Architect can cite a research brief in a design decision. See Research a question, which also covers research projects.

Questions

Section titled “Questions”

How do I pick between a quick answer and a project? Use one session for a look-up, a fact-check, or a brief. Start a project when the question needs days or a growing pile of sources.

Can I trust the citations? Each claim carries a grade and its sources, and the devil’s advocate pass looks for counter-evidence. Check any claim that matters against its source.

Does it save my research into project memory? Not unless you choose to. Research keeps its own sources and verdicts. After a finished repository survey or project synthesis, it may offer a reusable lesson to project-knowledge.

Does it edit my code? No. The retrieval subagents are read-only.

Details

Section titled “Details”

Scope: user (default), repo · Version: 1.1.10 · Source: packs/desk-research