Deep Research Skills

Deep Research Skills is an open-source set of skills and agents for Claude Code, OpenCode, and Codex, published by Weizhena on GitHub under the MIT license. It supports a two-phase research workflow: generating an extensible research outline, then investigating each item in depth. The README says it is inspired by the RhinoInsight paper on control mechanisms for deep research.

What It Is

The project is a collection of slash-command skills plus a web-search agent that runs inside coding-agent environments. A user supplies a topic, the skill proposes a list of items to research and the fields to collect for each, and the user can review and extend that outline before any deep investigation begins. The README lists use cases in academic research, technical research, market research, and due diligence.

How the Workflow Runs

• /research generates an outline of items and fields for a topic.
• /research-add-items and /research-add-fields extend an existing outline.
• /research-deep researches each item with parallel agents that search the web.
• /research-report turns the JSON results into a markdown report with a table of contents.

In the README's example, researching "AI Agent Demo 2025" produces an outline of 17 agents with fields to collect, followed by per-item details and a final report.md.

Setup Path

Installation is done by cloning the repository and copying the skills, agent definitions, and search modules into the configuration directory of the chosen tool (Claude Code, OpenCode, or Codex). English and Chinese skill versions are provided. PyYAML is required. OpenCode needs the OPENCODE_ENABLE_EXA=1 environment variable for web search, and Codex needs multi-agent settings in its config.toml, which an install script can apply.