Multi-LLM parallel research with query decomposition and synthesis
Uses parallel LLM agents to conduct exhaustive, discovery-driven research with query decomposition and synthesis. Activates when you need current, multi-perspective information on complex topics like travel deals, credit card perks, or market analysis.
/plugin marketplace add krishagel/geoffrey/plugin install geoffrey@geoffreyThis skill is limited to using the following tools:
RESEARCH_ENHANCEMENT.mdagents/claude-researcher.mdagents/gemini-researcher.mdagents/openai-researcher.mdagents/perplexity-researcher.mdcommands/conduct-research.mdscripts/context-loader.jsscripts/orchestrator.jsDiscovery-driven, exhaustive research using parallel LLM agents for comprehensive, current information gathering.
CRITICAL: These principles define how research works.
Before diving into research, ask deep, thought-provoking questions to understand:
Example: "Research Japan ski trip" should prompt:
Research automatically loads relevant context from preferences based on detected domain.
research_domains.[domain] from preferencesFor travel research, use browser-control to fetch current:
Use this skill when user needs:
CRITICAL: Travel research MUST include user-generated content:
Use the browser-control skill to scrape full threads and reviews that require:
Prerequisites:
cd skills/browser-control && ./scripts/launch-chrome.shAvailable Scripts (in skills/browser-control/scripts/):
# Navigate and get page content
bun navigate.js "https://www.reddit.com/r/JapanTravel/comments/..."
# Extract specific content
bun extract.js "https://flyertalk.com/forum/thread" ".post-content" --all
# Search travel sites directly
bun search.js reddit "Hakuba ski resort tips"
bun search.js flyertalk "Marriott Bonvoy Japan redemption"
# Capture current page without navigating
bun capture.js /tmp/screenshot.png
When to Use Browser Control:
Start with WebSearch/WebFetch for:
Escalate to browser-control when:
Example Flow:
WebSearch for blog reviewsbrowser-control to get full thread contentCRITICAL: All research output MUST include citations.
Every factual claim must link to its source:
Every research report must end with a complete "Sources" section listing all URLs used.
If you can't cite a source for a claim, either:
Break complex questions into 3-7 sub-queries covering different angles:
Launch multiple researcher agents simultaneously:
Agents are in ./agents/ directory:
| Agent | Best For | API Required |
|---|---|---|
perplexity-researcher.md | Current web info, citations | PERPLEXITY_API_KEY |
gemini-researcher.md | Multi-angle comparison | GEMINI_API_KEY |
openai-researcher.md | Structured analysis | OPENAI_API_KEY |
claude-researcher.md | Deep reasoning | Native (Claude Code) |
User: "Research the best ways to maximize Alaska Airlines miles for Japan flights"
Geoffrey:
1. Decomposes into sub-queries:
- Current Alaska redemption rates to Japan
- Partner airline options (JAL, etc.)
- Sweet spots and award availability patterns
- Credit card earning strategies
- Recent devaluations or changes
2. Launches 4 agents in parallel
3. Synthesizes findings with confidence scores
Use /conduct-research [topic] to trigger full parallel research workflow.
## Research: [Topic]
### Context Applied
- Domain: [travel/shopping/work/etc]
- User context loaded: [what was loaded from preferences]
- Dynamic data fetched: [current point balances, etc]
### Executive Summary
[2-3 paragraph overview of key findings and main recommendation]
### Detailed Findings
#### Sub-topic 1
Information with citations throughout. For example, "The Westin Rusutsu
costs 70,000 points peak" ([Point Hacks](url)) and "offers ski-in/ski-out
access" ([Marriott](url)).
#### Sub-topic 2
More information with inline citations...
### Multimedia Resources
- [Video: Title](url) - brief description
- [Podcast: Episode](url) - brief description
- [TikTok: @user](url) - brief description
### What I Discovered You Might Not Know
- Surprising finding 1 that user likely didn't know
- Surprising finding 2
- New source or perspective discovered
### Recommendations
1. Actionable recommendation specific to user's context
2. Another action based on their situation
### Confidence Assessment
- High confidence: [topics with multiple agreeing sources]
- Needs verification: [topics with limited sources]
- Conflicting information: [where sources disagree]
### All Sources
Complete list of every URL referenced:
- [Source Name 1](full-url)
- [Source Name 2](full-url)
- [Source Name 3](full-url)
Note: Every factual claim needs an inline citation. The "All Sources" section at the end provides a consolidated reference list.
ALWAYS save research reports to Google Docs after completing research.
cd skills/google-workspace
bun docs/create_doc.js <account> --title "<Research Topic>" --folder Research --content "<markdown>"
Will also save to local Obsidian vault (when skill is built).
Keys stored in: ~/Library/Mobile Documents/com~apple~CloudDocs/Geoffrey/secrets/.env
PERPLEXITY_API_KEY=pplx-...
GEMINI_API_KEY=...
OPENAI_API_KEY=sk-...
# Note: Claude runs natively in Claude Code, no API key needed
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.
Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.