Research Skill

This skill delegates multi-source research requests to an isolated worker with access to configured external sources and the local Obsidian vault.

[!important] Worker, not Subagent This skill uses a Worker (isolated claude -p process), not the Task Tool. Use Bash to execute run-worker.sh, NOT the Task Tool.

When to Use This Skill

Activate when the user wants information from multiple sources:

•"Find all information about X in Jira and Confluence"
•"Search for Y in all sources"
•"What has been communicated about Z?"
•"Research topic X across all available sources"
•"Gather context about feature Y from all available sources"

Single-source requests → Use specialized tools/workers instead:

•Issue tracker only → jira-reader skill
•Vault only → mcp__obsidian__search_vault_smart

Available Sources

Source	Type	Access
Obsidian Vault	Local knowledge base	`mcp__obsidian__*`
Issue Tracker	Issues, epics, comments	Configured MCP
Documentation	Wiki pages	Configured MCP
Chat	Conversations, threads	Configured MCP

Workflow

Step 1: Analyze Request

Identify:

•Topic/Keywords - What to search for
•Sources - Which sources to include (default: all)
•Time scope - If relevant (e.g., "last 2 weeks")
•Output format - Summary, detailed report, or link collection

Step 2: Formulate Query

Translate the user request into a research task:

User Request	Query for Worker
"Find info about feature X in Jira and Confluence"	"Research feature X: search issue tracker for related issues and documentation"
"What did team communicate about topic Y in Slack?"	"Search chat for topic Y discussions in the last 2 weeks"
"Gather all context about feature Z"	"Research feature Z across all sources: Vault, issues, documentation, chat"

Step 3: Invoke Worker

Start the isolated Research Worker via Bash (NOT Task Tool!):

bash

.claude/workers/.shared/run-worker.sh research "<research-task>"

Important: Use run_in_background: true and monitor with TaskOutput.

Step 4: Present Result

The worker outputs a structured research report. Present to the user:

•Summary of findings
•Key sources with links
•Open questions if information was incomplete

Example Queries

bash

# Multi-source research
.claude/workers/.shared/run-worker.sh research "Research feature X implementation: search issues for related tickets, documentation for technical docs, and chat for recent discussions"

# Issue tracker + Documentation
.claude/workers/.shared/run-worker.sh research "Find all information about pricing feature in issues and documentation"

# All sources
.claude/workers/.shared/run-worker.sh research "Gather comprehensive context about login feature from Vault, issues, documentation, and chat"

# Time-scoped
.claude/workers/.shared/run-worker.sh research "Search chat for discussions about topic Y since last month"

Research Hierarchy

The worker follows this priority:

•Obsidian Vault - Check local knowledge first
•Issue Tracker - Issues, epics, comments for project context
•Documentation - Specs and documentation
•Chat - Recent discussions and decisions

Output Format

The worker returns a structured markdown report:

markdown

# Research: [Topic]

**Date:** YYYY-MM-DD
**Status:** Complete | Partial | Insufficient

## Summary
[2-3 sentence executive summary]

## Key Findings
- Finding 1
- Finding 2

## Sources

### Issues
- [PROJECT-XXX](url) - Description

### Documentation
- [Page Title](url) - Description

### Chat
- [Thread](url) - Key discussion point

## Open Questions
- Unanswered question 1

Technical Details

•Worker: .claude/workers/research/
•MCP: Configured sources (Obsidian, Atlassian, Slack, etc.)
•Permissions: Read-only for all sources
•Output: Direct stdout (structured markdown)