123 lines
3.1 KiB
Markdown
123 lines
3.1 KiB
Markdown
# Conduct Research Command
|
|
|
|
Orchestrate parallel multi-LLM research on a topic.
|
|
|
|
## Usage
|
|
```
|
|
/conduct-research [topic or question]
|
|
```
|
|
|
|
## Workflow
|
|
|
|
### Step 1: Query Decomposition
|
|
|
|
Break the research topic into 3-7 focused sub-queries. Consider:
|
|
- **Factual**: What are the basic facts?
|
|
- **Current**: What's the latest information?
|
|
- **Comparative**: How does it compare to alternatives?
|
|
- **Practical**: How does it work in practice?
|
|
- **Optimization**: What are the best strategies?
|
|
- **Gotchas**: What are common mistakes or pitfalls?
|
|
|
|
Example for "maximize Alaska Airlines miles for Japan":
|
|
1. Current Alaska redemption rates to Japan partners
|
|
2. JAL vs Cathay vs other OneWorld options for Japan routes
|
|
3. Best credit card strategies for earning Alaska miles
|
|
4. Award availability patterns and booking tips
|
|
5. Recent program changes or devaluations
|
|
|
|
### Step 2: Parallel Agent Execution
|
|
|
|
Launch researcher agents in parallel using the Task tool. Each agent gets relevant sub-queries:
|
|
|
|
**Perplexity Agent** (current info)
|
|
- Sub-queries about recent changes, current rates, latest news
|
|
|
|
**Gemini Agent** (comparisons)
|
|
- Sub-queries comparing options, trade-offs
|
|
|
|
**OpenAI Agent** (structure)
|
|
- Sub-queries about processes, frameworks, how things work
|
|
|
|
**Claude Agent** (synthesis)
|
|
- Sub-queries about recommendations, implications
|
|
|
|
Use this pattern:
|
|
```
|
|
Launch 4 Task agents in parallel:
|
|
- Task: perplexity-researcher with [sub-queries 1, 2]
|
|
- Task: gemini-researcher with [sub-queries 3, 4]
|
|
- Task: openai-researcher with [sub-query 5]
|
|
- Task: claude-researcher with [synthesis query]
|
|
```
|
|
|
|
### Step 3: Result Collection
|
|
|
|
Gather results from all agents (typically 15-45 seconds).
|
|
|
|
### Step 4: Synthesis
|
|
|
|
Combine all findings into a cohesive report:
|
|
|
|
1. **Identify Consensus** - What do multiple sources agree on?
|
|
2. **Flag Conflicts** - Where do sources disagree?
|
|
3. **Score Confidence** - High (4/4 agree), Medium (3/4), Low (2/4 or conflicting)
|
|
4. **Extract Actions** - What should the user actually do?
|
|
|
|
### Step 5: Final Report
|
|
|
|
Format output as:
|
|
|
|
```markdown
|
|
## Research Report: [Topic]
|
|
|
|
### Executive Summary
|
|
[2-3 sentence overview of key findings]
|
|
|
|
### Key Findings
|
|
|
|
#### High Confidence (Multiple Sources Agree)
|
|
- Finding 1
|
|
- Finding 2
|
|
|
|
#### Medium Confidence
|
|
- Finding 3
|
|
- Finding 4
|
|
|
|
#### Needs Verification
|
|
- Finding 5 (sources conflict)
|
|
|
|
### Detailed Analysis
|
|
[Organized by sub-topic with source attribution]
|
|
|
|
### Sources
|
|
- [Source 1](url) - via Perplexity
|
|
- [Source 2](url) - via Gemini
|
|
- etc.
|
|
|
|
### Recommendations
|
|
1. **Do This First**: [Action]
|
|
2. **Then Consider**: [Action]
|
|
3. **Watch Out For**: [Gotcha]
|
|
|
|
### Research Metadata
|
|
- Agents used: Perplexity, Gemini, OpenAI, Claude
|
|
- Time: [X seconds]
|
|
- Date: [Today's date]
|
|
- Confidence: Overall High/Medium/Low
|
|
```
|
|
|
|
## API Requirements
|
|
|
|
Ensure these are configured in `~/Library/Mobile Documents/com~apple~CloudDocs/Geoffrey/secrets/.env`:
|
|
- PERPLEXITY_API_KEY
|
|
- GEMINI_API_KEY
|
|
- OPENAI_API_KEY
|
|
|
|
## Notes
|
|
|
|
- If an API fails, continue with remaining agents
|
|
- Note which agents were used in final report
|
|
- For time-sensitive topics, prioritize Perplexity results
|
|
- Consider user preferences when making recommendations (check knowledge-manager)
|