karvifi/os/auto-research

SKILL: Auto Research

Purpose

Conduct comprehensive research autonomously — from question to actionable findings — without needing constant direction.

Process

Phase 1: Question Decomposition

Break the research question into sub-questions:

• Core question (what needs to be known)
• Context questions (background needed)
• Comparative questions (what alternatives exist)
• Gap questions (what is unknown or unsettled)

Phase 2: Multi-Source Intelligence Gathering

Primary sources (use MCP tools):

brave-search MCP:
- "[topic] research 2024 2025"
- "[topic] state of the art"
- "[topic] best practices"
- "[topic] common mistakes"
- "[topic] vs [alternative]"
 
github MCP:
- Search repos: gh search repos [topic] --sort stars
- Find implementations: gh search code [topic]
 
fetch MCP:
- Official documentation
- Academic papers (arXiv, PubMed, Semantic Scholar)
- Industry reports
 
context7 MCP:
- Official library/framework documentation

Information triage:

• Tier 1: Primary sources (official docs, peer-reviewed, original research)
• Tier 2: Secondary sources (established blogs, reputable news)
• Tier 3: Community sources (Stack Overflow, Reddit, forums)

Phase 3: Synthesis

For each sub-question, produce:

• What is definitively known (with sources)
• What is contested (multiple valid views)
• What is unknown (gaps)
• What this means for the project/decision

Phase 4: Actionable Output

Convert research into decisions:

• What should we USE? (with reasoning)
• What should we AVOID? (with reasoning)
• What needs MORE RESEARCH? (specific questions)
• What is the RECOMMENDATION? (concrete)

Output Template

# Research Report: [Topic]
Date: [date]
Sources: [count] primary, [count] secondary
 
## Executive Summary (3 sentences max)
[What was found + key recommendation]
 
## Key Findings
### [Finding 1 — most important]
Evidence: [source]
Confidence: High / Medium / Low
 
### [Finding 2]
...
 
## Competitive Landscape (if applicable)
| Option | Pros | Cons | Best for |
|--------|------|------|---------|
 
## Recommendation
**Use: [X]**
Because: [specific reasons tied to the project context]
 
## What Needs More Research
-[specific open question]
 
## Sources
-[URL]: [what it provided] — [date]

Quality checks

• [ ] Every claim has a source
• [ ] No fabricated data or statistics
• [ ] Contradicting sources acknowledged
• [ ] Recommendation is specific, not vague
• [ ] Research is recent (within 6 months for fast-moving fields)