Research Brief
Turns an open question into a sourced, skimmable brief. Decomposes the question, builds a search plan that deliberately looks for disconfirming evidence (not just confirming), grades every source, tracks every claim to its source(s) in a claim ledger, and writes up findings with honest confidence levels — distinguishing well-corroborated claims from single-source or disputed ones.
When to use
- Answering an open-ended "should we / is it true that / what's the state of X" question that deserves more than an off-the-cuff answer.
- Producing something shareable — a brief a reader can check, not just a paragraph of assertions.
- Any research task where being wrong confidently is worse than being appropriately uncertain (technical decisions, due diligence, fact-checking).
Install
npx openagents-cli add openagents/research-brief| Runtime | Installed to |
|---|---|
claude-code | .claude/skills/research-brief/ |
cursor | .cursor/rules/research-brief/ |
codex | .codex/skills/research-brief/ |
openai-agents | .openai-agents/research-brief/ |
generic | .openagents/research-brief/ |
Inputs
| name | type | required | default | description |
|---|---|---|---|---|
question | string | yes | — | The question or claim to research |
depth | string | no | standard | quick (3-5 sources), standard (6-12), deep (12+, diverse source types) |
audience | string | no | — | Who the brief is for, to calibrate depth/framing |
Example run
> Write a research brief on whether we should adopt library X over Y for our queue.
The agent decomposes the question (capability differences, maintenance health, known
limitations, migration cost), searches for both supporting and disconfirming
evidence per sub-question, grades each source (source-grading.md), builds a claim
ledger, and writes the result with templates/brief.md — leading with a direct
bottom-line answer and an honest confidence rating.
Files
SKILL.md— the decompose/search/grade/ledger/write procedure (entry point).source-grading.md— the A–F source reliability scale and rules for using it.templates/brief.md— the brief output format.
Limitations
- Quality depends on what's actually searchable/accessible in the runtime environment; paywalled or non-indexed primary sources may be represented only via secondary citations, noted as such.
- Does not independently verify claims beyond cross-referencing sources — it is not a substitute for expert review on highly technical or high-stakes questions.
deepruns take meaningfully longer due to the larger source count and deliberate type diversity requirement.