deep-research-skill

A Cursor Agent Skill for deep, multi-step AI research. Decomposes any query into a parallel subagent execution plan, accumulates structured evidence with confidence ratings, and delivers the right artifact type rather than a wall of text.

Install

The SKILL.md format is compatible with both Cursor and Claude Code. Install path differs by harness.

Cursor - personal (all projects):

mkdir -p ~/.cursor/skills/deep-research
curl -o ~/.cursor/skills/deep-research/SKILL.md \
  https://raw.githubusercontent.com/TheAdamLabs/deep-research-skill/main/SKILL.md

Cursor - project (shared via .cursor/skills/):

mkdir -p .cursor/skills/deep-research
curl -o .cursor/skills/deep-research/SKILL.md \
  https://raw.githubusercontent.com/TheAdamLabs/deep-research-skill/main/SKILL.md

Claude Code - personal (all projects):

mkdir -p ~/.claude/skills/deep-research
curl -o ~/.claude/skills/deep-research/SKILL.md \
  https://raw.githubusercontent.com/TheAdamLabs/deep-research-skill/main/SKILL.md

Claude Code - project (shared via .claude/skills/):

mkdir -p .claude/skills/deep-research
curl -o .claude/skills/deep-research/SKILL.md \
  https://raw.githubusercontent.com/TheAdamLabs/deep-research-skill/main/SKILL.md

Start a new chat to pick up the skill. No restart needed in Claude Code (skills hot-reload).

Note: The parallel subagent phase requires harness support for spawning subagents. In Cursor this uses the Task tool. In Claude Code it uses claude -p via bash. The skill body uses generic language; the agent resolves to whatever its harness provides.

Usage

Use the deep-research skill to compare pricing models for B2B SaaS infrastructure tools

The skill decides what files to produce based on query complexity. Results are saved to a timestamped directory under research/; chat output is always a short summary block pointing to the files.

Architecture

Main agent: query analysis -> DAG plan -> plan critique
    |
    +-- [parallel] Subagent A -> evidence-A.md
    +-- [parallel] Subagent B -> evidence-B.md
    +-- [parallel] Subagent C -> evidence-C.md
         |
         +-- [depends on A] Subagent D -> evidence-D.md
    |
Main agent: merge evidence -> gap analysis
    |
    +-- [parallel, if gaps] Gap subagent(s)
    |
Main agent: artifact -> self-score (0-12)
    |
    +-- [if score 7-9] Patch subagent -> re-score
    |
Main agent: dashboard (if warranted) -> summary to chat

License

MIT