Deep Research Skill

Overview

Deep Research Skill is a structured research workflow for Claude Code, OpenCode, and Codex, inspired by the RhinoInsight paper on control mechanisms for deep research. Instead of one open-ended search pass, it splits research into an editable outline phase and a deep-investigation phase, so you can correct the research plan (add items, add fields) before agents spend time searching. It targets academic surveys, technology/framework comparisons, market and competitor research, and due-diligence-style company research.

How it works

1. /research <topic> generates an outline: a list of items to research (e.g. 17 AI agents) plus the fields to collect for each (company, pricing, tech specs, reviews...).
2. /research-add-items / /research-add-fields let you extend the outline before committing to the expensive phase, if the first draft missed items or data points.
3. /research-deep dispatches parallel web-search agents, one per item, each populating the fields defined in the outline.
4. /research-report turns the collected JSON results into a single markdown report with a table of contents.
5. A separate web-search-agent (installed alongside the skill) does the actual searching; on OpenCode it requires OPENCODE_ENABLE_EXA=1 to get real web search instead of the weaker built-in web fetch.

Examples

/research AI Agent Demo 2025
→ outline: 17 AI agents to research (ChatGPT Agent, Claude Computer Use,
  Cursor, etc.) with fields like company, release date, pricing, tech specs

/research-deep
→ searches the web for each item in parallel, filling in every field

/research-report
→ report.md: a complete markdown report with a table of contents

Installation

You need Python and pip install pyyaml; Claude Code 2.1.0+ supports triggering /research directly, older versions need run /research.

1. Clone the repo: git clone https://github.com/Weizhena/deep-research-skills.git && cd deep-research-skills.
2. Copy the skill files: cp -r skills/research-en/* ~/.claude/skills/ (use research-zh for the Chinese version, or the research-codex-en / research-codex-zh folders into ~/.codex/skills/ for Codex).
3. Install the required web-search agent: cp agents/web-search-agent.md ~/.claude/agents/ and cp -r agents/web-search-modules ~/.claude/agents/.
4. Install the Python dependency: pip install pyyaml.
5. On OpenCode only, also run export OPENCODE_ENABLE_EXA=1 (add it to ~/.bashrc to persist) so web search actually hits the network instead of falling back to plain web fetch.
6. Verify it worked: run /research <a topic you care about> and confirm you get back an outline of items and fields before anything is searched.