Research

⚠️ MANDATORY TRIGGER

When user says "research" (in any form), ALWAYS invoke this skill.

"Research" alone = Standard mode. No exceptions.

Customization

Before executing, check for user customizations at: ~/.claude/PAI/USER/SKILLCUSTOMIZATIONS/Research/

If this directory exists, load and apply any PREFERENCES.md, configurations, or resources found there. These override default behavior. If the directory does not exist, proceed with skill defaults.

2. Output text notification:

   Running the **WorkflowName** workflow in the **Research** skill to ACTION...
   

This is not optional. Execute this curl command immediately upon skill invocation.

Research Skill

Comprehensive research, analysis, and content extraction system.

MANDATORY: URL Verification

READ: UrlVerificationProtocol.md - Every URL must be verified before delivery.

Research agents hallucinate URLs. A single broken link is a catastrophic failure.

Workflow Routing

Route to the appropriate workflow based on the request.

CRITICAL: For due diligence, company/person background checks, or vetting -> INVOKE OSINT SKILL INSTEAD

Research Modes (Primary Workflows)

• Quick/minor research (1 Perplexity, 1 query) -> Workflows/QuickResearch.md
• Standard research - DEFAULT (3 agents: Perplexity + Claude + Gemini) -> Workflows/StandardResearch.md
• Extensive research (4 types x 3 threads = 12 agents) -> Workflows/ExtensiveResearch.md
• Deep investigation / iterative research (progressive deepening, loop-compatible) -> Workflows/DeepInvestigation.md

Deep Content Analysis

• Extract alpha / deep analysis / highest-alpha insights -> Workflows/ExtractAlpha.md

Content Retrieval

• Difficulty accessing content (CAPTCHA, bot detection, blocking) -> Workflows/Retrieve.md
• YouTube URL extraction (use fabric -y URL immediately) -> Workflows/YoutubeExtraction.md
• Web scraping -> Workflows/WebScraping.md

Specific Research Types

• Claude WebSearch only (free, no API keys) -> Workflows/ClaudeResearch.md
• Perplexity API research (use Quick for single-agent) -> Workflows/QuickResearch.md
• Interview preparation (Tyler Cowen style) -> Workflows/InterviewResearch.md
• AI trends analysis -> Workflows/AnalyzeAiTrends.md

Fabric Pattern Processing

• Use Fabric patterns (242+ specialized prompts) -> Workflows/Fabric.md

Content Enhancement

• Enhance/improve content -> Workflows/Enhance.md
• Extract knowledge from content -> Workflows/ExtractKnowledge.md

Research Index — Institutional Knowledge

All research workflows MUST use the Research Index (PAI/Tools/ResearchIndex.ts):

1. Before research: Run bun ~/.claude/PAI/Tools/ResearchIndex.ts dedup "<topic>" to check for prior findings
2. After research: Run bun ~/.claude/PAI/Tools/ResearchIndex.ts save --topic "<topic>" --summary "<findings>" ... to capture results
3. Context seeding: When prior research exists, include its summary in agent prompts so they build on (not repeat) earlier work

CLI reference:

bun ~/.claude/PAI/Tools/ResearchIndex.ts query "topic"   # Search prior research
bun ~/.claude/PAI/Tools/ResearchIndex.ts dedup "topic"   # Check for recent duplicates
bun ~/.claude/PAI/Tools/ResearchIndex.ts save --topic "t" --summary "s" --agents "a,b" --keywords "k1,k2" --quality N
bun ~/.claude/PAI/Tools/ResearchIndex.ts list             # List recent entries
bun ~/.claude/PAI/Tools/ResearchIndex.ts stats            # Index statistics

Quick Reference

READ: QuickReference.md for detailed examples and mode comparison.

Integration

Feeds Into

• blogging - Research for blog posts
• newsletter - Research for newsletters
• xpost - Create posts from research

Uses

• be-creative - deep thinking for extract alpha
• OSINT - MANDATORY for company/people comprehensive research
• BrightData MCP - CAPTCHA solving, advanced scraping
• Apify MCP - RAG browser, specialized site scrapers

Deep Investigation Mode

Progressive iterative research that builds a persistent knowledge vault. Works in both single-run (one cycle) and loop mode (Algorithm-driven iterations).

Concept: Broad landscape → discover entities → score importance/effort → deep-dive one at a time → loop until coverage complete.

Domain template packs customize the investigation for specific domains:

• Templates/MarketResearch.md — Companies, Products, People, Technologies, Trends, Investors
• Templates/ThreatLandscape.md — Threat Actors, Campaigns, TTPs, Vulnerabilities, Tools, Defenders
• No template? The workflow creates entity categories dynamically from the landscape research.

Example invocation:

"Do a deep investigation of the AI agent market"
→ Loads MarketResearch.md template
→ Iteration 1: Broad landscape + first entity deep-dive
→ Loop mode: Each iteration deep-dives the next highest-priority entity
→ Exit: When all CRITICAL/HIGH entities researched + all categories covered

Artifacts persist at ~/.claude/MEMORY/RESEARCH/{date}_{topic}/ — the vault survives across sessions.

See Workflows/DeepInvestigation.md for full workflow details.

File Organization

Working files (temporary work artifacts): ~/.claude/MEMORY/WORK/{current_work}/

• Read ~/.claude/MEMORY/STATE/current-work.json to get the work_dir value
• All iterative work artifacts go in the current work item directory
• This ties research artifacts to the work item for learning and context

History (permanent): ~/.claude/History/research/YYYY-MM/YYYY-MM-DD_[topic]/