research（研究）

/research - Intelligent Research Orchestration

I'll coordinate specialized agents to conduct systematic research on your topic.

Research Topic

Query: $ARGUMENTS

🤖 Orchestration Mode

I'll analyze your query and build an execution plan using specialized agents.

Agent Registry:

• literature-reviewer: Academic source search, citation chains
• critical-analyzer: Fallacy detection, bias identification
• hypothesis-explorer: Hypothesis formulation, variable mapping
• lateral-thinker: Cross-domain analogies, creative thinking
• qual-researcher: Thematic analysis, coding strategies
• quant-analyst: Statistical methods, effect sizes
• peer-reviewer: Manuscript evaluation
• ethics-expert: IRB compliance, privacy risks
• grant-writer: Grant proposal development, funding strategy

Orchestration Process:

1. Classify Query: Determine research type
2. Select Agents: Choose optimal agent sequence
3. Generate Plan: Build execution DAG
4. Present Plan: Show workflow to user
5. Execute: Run agents sequentially, save outputs
6. Synthesize: Integrate outputs, save final report

I'm analyzing your research topic to determine the optimal workflow...

Proposed Plan: [I will generate this based on your query]

Proceed? (yes/no/modify)

File Writing Protocol

CRITICAL: After plan approval, you MUST save all research outputs to files.

Step 1: Create Output Directory

After user approves the plan, immediately:

1. Use Bash to create timestamped directory: mkdir -p research-outputs/$(date +%Y-%m-%d_%H-%M-%S)
2. Store the full path in a variable for subsequent file writes
3. Confirm directory creation to user

Step 2: Write Research Plan

Use Write tool to create 00-research-plan.md in the output directory:

# Research Plan: [Query]

**Created**: [Timestamp]
**Query**: [User's research query]

## Selected Agents
1. [agent-name] - [purpose]
2. [agent-name] - [purpose]
...

## Execution Plan
[Full plan as presented to user]

Step 3: Write Agent Outputs

After EACH agent completes execution:

1. Use Write tool to save output to {NN}-{agent-name}.md

- NN = sequential number (01, 02, 03, etc.) - Example: 01-literature-reviewer.md, 02-critical-analyzer.md

1. Format:

# {Agent Name} Output

**Agent**: {agent-name}
**Executed**: [Timestamp]

---

[Full agent output - preserve all markdown formatting]

Step 4: Write Final Synthesis

After synthesis completes:

1. Use Write tool to save to final-synthesis.md
2. Include complete synthesis with all sections
3. Add metadata header with timestamp and agents used

Error Handling

• If directory creation fails: warn user and continue with conversation-only output
• If Write fails: log error, notify user, continue execution
• Partial results are acceptable if execution is interrupted

Modes Supported:

• Default: Interactive
• --auto: Automatic execution
• --plan-only: Show plan only
• --manual: Traditional guided mode

Templates: --template=quick|rigorous|comprehensive