deep-research
Conduct deep web research using the openbrowser-ai agent: decompose a query, investigate sub-questions across multiple sources, and produce a cited markdown report plus structured JSON under local_docs/research/. Trigger when the user asks to: research a topic, do a deep dive, investigate, gather evidence, compare options, write a literature review, build a briefing, or produce a cited report.
npx skills add majiayu000/claude-skill-registry --skill deep-research --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
Drives openbrowser-ai to research a topic by decomposing the query into sub-questions, dispatching parallel sub-agents (one tab per sub-question) to gather findings, and merging results into a cited markdown report plus a JSON file under local_docs/research. Supports flat synthesis (default) and drilldown mode detected from prompt wording. Outputs to local_docs/research/YYYY-MM-DD-<slug>.md and .json, with per-sub-agent JSONs under local_docs/research/_partial/agent-<NN>.json. Uses a dedicated orchestrator that dispatches parallel agents via /dispatching-parallel-agents and ensures each sub-agent owns exactly one tab. If a first wave yields fewer than 2 findings, the workflow may trigger Step 2b retry sub-agents. All facts include citations and adhere to the tool and file naming conventions specified in the skill.
How it works
- The orchestrator decomposes the user query into 3-7 sub-questions and selects mode: drilldown if the query matches drilldown-like terms; otherwise flat.
- It creates output paths under local_docs/research/, ensuring the date-based slug is unique by appending suffixes as needed.
- It dispatches one sub-agent per sub-question through /dispatching-parallel-agents, with each sub-agent owning exactly one tab and never opening new tabs. Each sub-agent writes findings to local_docs/research/_partial/agent-<AGENT_INDEX>.json.
- After all sub-agents return, the orchestrator merges per-agent findings into a single report. If any first-wave sub-agent returns fewer than 2 findings, a Step 2b retry is triggered for weak sub-questions with broader search strategies, using agent-retry-<AGENT_INDEX>.json files.
- Retry phase uses the same one-tab discipline and writes to local_docs/research/_partial/agent-retry-<AGENT_INDEX>.json; results are merged back into the overall findings.
- The final output includes a markdown report and a structured JSON file under local_docs/research/ with the same date-based slug.
When to use it
Trigger when the user asks to research a topic, do a deep dive, investigate, gather evidence, compare options, write a literature review, build a briefing, or produce a cited report.
What it can touch
- local_docs/research/ (for outputs and the date-based slug)
- local_docs/research/_partial/ (for per-sub-agent JSONs, agent-<NN>.json and agent-retry-<NN>.json)
- /dispatching-parallel-agents endpoint to spawn sub-agents
Caveats
- Requires openbrowser-ai daemon; step-by-step session management and tab isolation are mandated. All citations appear in the report as footnotes [N].
- If a first-wave sub-agent returns fewer than 2 findings, Step 2b may be invoked with retry sub-agents. The retry path prohibits -p mode and continues to use the same daemon with one tab per sub-agent.
- The orchestrator never drives tabs itself; it plans, dispatches, merges, renders, verifies, and cleans up.
# Deep Research Drive `openbrowser-ai` to investigate a topic across multiple web sources and produce a cited markdown report plus structured JSON. Two modes: - **flat synthesis** (default) -- decompose query into 3-7 sub-questions, dispatch one parallel sub-agent per sub-question (each owns one tab), merge into one cited report. - **drilldown** (auto-detected from prompt phrasing: "deep dive", "exhaustive", "recursive", "drilldown", "thorough") -- same as flat, plus a second wave of parallel sub-agents on findings flagged `needs_depth=true`. Hard cap depth=2, max 3 follow-up sub-agents per parent. Output paths (relative to current project root): - `local_docs/research/YYYY-MM-DD-<slug>.md` - `local_docs/research/YYYY-MM-DD-<slug>.json` **Architecture (mandatory):** the orchestrating Claude session (the one running this skill) MUST dispatch parallel sub-agents via `/dispatching-parallel-agents`, one sub-agent per sub-question. Each sub-agent owns exactly ONE tab. Sub-agents do not open additional tabs. The orchestrator merges per-agent findings into one report. Why one tab per sub-agent and not `asyncio.gather` over tabs in a single `-c` call: a single Python coroutine driving N ta
- Setup
- Workflow
- Step 0 -- Session check
- Step 1 -- Plan
- Step 2a -- Dispatch parallel sub-agents (one tab per agent)
- Step 2b -- Retry weak sub-questions with broader sub-agents
- Step 3 -- Drilldown (drilldown mode only): second wave of parallel sub-agents
- Step 4 -- Aggregate and dedup
- Step 5 -- Render report
- Step 6 -- Verify citations
- Step 7 -- Cleanup
- Tips
- Cleanup
openbrowser-ai --help macOS / Linux curl -fsSL https://openbrowser.me/install.sh | sh Windows PowerShell irm https://openbrowser.me/install.ps1 | iex export OPENBROWSER_HEADLESS=true mkdir -p local_docs/research if openbrowser-ai daemon status 2>&1 | grep -qi 'running\|listening\|pid'; then echo "Reusing existing daemon -- will work in new tabs" export DEEP_RESEARCH_REUSED=1
What does the deep-research skill do?
Conduct deep web research using the openbrowser-ai agent: decompose a query, investigate sub-questions across multiple sources, and produce a cited markdown report plus structured JSON under local_docs/research/. Trigger when the user asks to: research a topic, do a deep dive, investigate, gather evidence, compare options, write a literature review, build a briefing, or produce a cited report.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill deep-research --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.