Research Agent Skill

ct-research-agent - Multi-source research and investigation

Research Agent Skill

The research agent skill provides comprehensive research capabilities, gathering information from multiple sources and producing structured outputs.

Overview

Multi-Source

Web search, library docs, Reddit discussions

Structured Output

Markdown reports with citations

Task Linking

Connect research to CLEO tasks

Manifest Integration

Automatic MANIFEST.jsonl entries

Sources

The research agent integrates multiple data sources:

Source	Provider	Best For
Web Search	Tavily	Current information, news
Library Docs	Context7	Framework/library documentation
Reddit	Tavily	Community discussions, opinions
Direct URLs	Tavily Extract	Specific page content

Usage

General Research

cleo research "React 19 new features"

Library Docs

cleo research --library react "useEffect cleanup patterns"

Reddit Discussions

cleo research --reddit "CLEO task management" --subreddit programming

URL Extraction

cleo research --url "https://docs.example.com/guide"

Research Depth

Basic

5-10 sources, quick overview

Standard

10-15 sources (default)

Deep

15-25 sources, comprehensive

# Quick research
cleo research "topic" -d basic

# Comprehensive research
cleo research "topic" -d deep

Output Structure

Research outputs follow a consistent structure:

# Research: [Topic]

**Date**: 2026-01-20
**Sources**: 15
**Depth**: standard

## Executive Summary
Key findings in 3-5 sentences.

## Key Findings
1. Finding one with citation [1]
2. Finding two with citation [2]
...

## Sources
[1] https://example.com/article
[2] https://docs.library.com/guide
...

## Recommendations
- Actionable item 1
- Actionable item 2

Task Linking

Connect research to CLEO tasks:

# Link during research
cleo research "auth patterns" --link-task T005

# Link after research
cleo research link T005 research-id-123

Manifest Integration

Research outputs automatically create manifest entries:

{
  "id": "auth-patterns-2026-01-20",
  "file": "research-outputs/2026-01-20_auth-patterns.md",
  "title": "Authentication Pattern Research",
  "key_findings": [
    "JWT with refresh tokens is current best practice",
    "OAuth 2.0 PKCE recommended for public clients",
    "Session duration: 15 min access, 7 day refresh"
  ],
  "linked_tasks": ["T005"]
}

Subcommands

research init

Initialize the research outputs directory:

cleo research init

Creates claudedocs/research-outputs/ with MANIFEST.jsonl

research list

List research entries from manifest:

cleo research list
cleo research list --status complete --limit 10

research show

Show research entry details:

cleo research show <id>
cleo research show <id> --full  # Include file content

research inject

Get subagent injection template:

cleo research inject
cleo research inject --clipboard

Subagent Workflow

When using research in orchestrated workflows:

1. Inject Protocol

   Use cleo research inject to get the research protocol for subagents

2. Subagent Research

   Subagent performs research and writes to MANIFEST.jsonl

3. Query Results

   Parent agent queries via cleo research list and cleo research show

4. Read Summaries

   Read only key_findings, not full research files

Best Practices

Specific Queries

Narrow queries yield better results than broad topics

Link Tasks

Always link research to relevant tasks for traceability

Use Depth Appropriately

Basic for quick checks, deep for thorough investigation

Read Summaries First

Check key_findings before reading full reports

Related

Commands: research

Complete research command reference

Orchestrator

Coordinate research with other work