multi-agent-researcher
Skill11 starsChanged 11 months agoApache-2.0View on GitHub
Conduct comprehensive research on any topic by coordinating 2-4 specialized researcher agents in parallel, then synthesizing findings into a detailed report via mandatory report-writer agent delegation (Full workflow documentation at docs/workflows/research-workflow.md)
In this repository3 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- semantic-searchSkill
- spec-workflow-orchestratorSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. About 0 are common, 12 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈12
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
Do NOT invoke for:
- Distinctive: line 55
Use TodoWrite to track:
- Distinctive: line 87
Update TodoWrite as researchers complete.
- Very specific: line 90
Use Glob to confirm all files exist: files/research_notes/*.md
- Very specific: line 98
YOU DO NOT HAVE WRITE TOOL ACCESS: when this skill is active.
- Very specific: line 114
Use Glob to confirm all research notes exist: files/research_notes/*.md
- Distinctive: line 115
Verify count matches number of spawned researchers
- Distinctive: line 116
If any missing: investigate and complete before synthesis
- Distinctive: line 154
After spawning report-writer agent, wait for completion.
- Distinctive: line 155
Read all research notes
- Very specific: line 157
Write comprehensive report to files/reports/
- Distinctive: line 212
Always spawn researchers simultaneously
- Distinctive: line 213
Never sequential unless dependent
- Distinctive: line 222
Add insights beyond individual notes
- Distinctive: line 230
Contradictory Findings: Document all perspectives explicitly
- Distinctive: line 231
Unclear Query: Ask clarifying questions first
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Multi-Agent Research Coordinator
Purpose
Transform complex research questions into comprehensive reports by:
- Decomposing broad topics into 2-4 focused subtopics
- Spawning specialized researcher agents in parallel
- Synthesizing findings into cohesive final report
- Saving structured outputs for reference
When to Use
Auto-invoke when user asks:
- Search/Discovery: "Search what is [topic]", "Find information about [subject]", "Look up [technology]", "Discover [patterns]"
- Investigation: "Research [topic]", "Investigate [subject]", "Analyze [phenomenon]", "Study [field]", "Explore [domain]"
- Collection: "Gather information about [subject]", "Collect data on [topic]", "Compile resources for [area]"
- Learning: "Learn about [subject]", "Tell me about [topic]", "Dig into [technology]", "Delve into [concept]"
- Contextual: "What are the latest developments in [field]?", "Comprehensive analysis of [topic]", "Deep dive into [subject]", "State of the art in [domain]", "Best practices for [area]"
Do NOT invoke for:
- Simple factual questions ("What is the capital of France?")
- Decision evaluation ("Should I use X or Y?")
- Code-related tasks ("Debug this function", "Write a script")
Orchestration Workflow
Phase 1: Query Analysis & Decomposition
Step 1.1: Understand the Research Question Analyze user's query to identify core topic, scope, and intent.
Step 1.2: Decompose into Subtopics Break topic into 2-4 focused subtopics that are:
- Mutually exclusive (minimal overlap)
- Collectively exhaustive (cover whole topic)
- Independently researchable
- Together provide comprehensive coverage
Decomposition Patterns:
Temporal: Past → Current → Future Categorical: Category 1, 2, 3 Stakeholder: Technical → Business → Policy → User Problem-Solution: Problem → Solutions → Gaps → Future Geographic: Region A → Region B → Comparison
Step 1.3: Create Research Plan Use TodoWrite to track:
- [ ] Decompose query into subtopics
- [ ] Spawn researcher 1: [subtopic]
- [ ] Spawn researcher 2: [subtopic]
- [ ] Spawn researcher 3: [subtopic]
- [ ] Synthesize findings
- [ ] Save final reportPhase 2: Parallel Research Execution
Step 2.1: Spawn Researcher Agents in Parallel
For each subtopic, create a Task tool call with:
subagent_type: "researcher"
description: "Research {subtopic name}"
prompt: "Research the following subtopic in depth:
**Subtopic**: {Subtopic name}
**Context**: Part of research on '{original topic}'
**Focus**: {Specific guidance}
Conduct thorough web research, gather authoritative sources, extract key findings, and save results to files/research_notes/{subtopic-slug}.md"Critical: Spawn ALL researchers in parallel (multiple Task calls in same message), not sequentially.
Step 2.2: Monitor Completion Update TodoWrite as researchers complete.
Step 2.3: Verify All Complete
Use Glob to confirm all files exist: files/research_notes/*.md
Phase 3: Synthesis & Report Generation
⚠️ CRITICAL: ARCHITECTURAL ENFORCEMENT ACTIVE ⚠️
YOU DO NOT HAVE WRITE TOOL ACCESS when this skill is active. The allowed-tools frontmatter explicitly EXCLUDES the Write tool to enforce proper workflow delegation.
YOU CANNOT:
- ❌ Write synthesis reports yourself
- ❌ Create files in files/reports/ directory
- ❌ Bypass the report-writer agent
YOU MUST:
- ✅ Spawn report-writer agent via Task tool
- ✅ Delegate all synthesis and report writing to the agent
- ✅ Read the completed report and deliver to user
Step 3.1: Verify Research Completion
- Use Glob to confirm all research notes exist:
files/research_notes/*.md - Verify count matches number of spawned researchers
- If any missing: investigate and complete before synthesis
Step 3.2: Spawn Report-Writer Agent (MANDATORY)
This is the ONLY synthesis approach - there is no "Option A" or "Option B". You MUST use the report-writer agent because you lack Write tool permissions.
Task:
subagent_type: "report-writer"
description: "Synthesize research findings into comprehensive report"
prompt: "Synthesize research into comprehensive report:
**Original Question**: {user query}
**Subtopics Researched**: {list all subtopics}
**Notes Location**: files/research_notes/
## Your Tasks:
1. Read ALL research notes from files/research_notes/
2. Identify themes, patterns, and contradictions across notes
3. Synthesize findings into cohesive narrative
4. Cite sources from research notes
5. Add cross-cutting insights beyond individual notes
6. Save comprehensive report to files/reports/{topic-slug}_{timestamp}.md
## Report Structure:
- Executive Summary
- Key Findings (with evidence from research notes)
- Detailed Analysis by subtopic
- Cross-Cutting Themes
- Contradictions and Debates
- Gaps and Limitations
- Source Bibliography
Use the timestamp format: $(date +\"%Y%m%d-%H%M%S\") for the filename."Step 3.3: Monitor Agent Completion
After spawning report-writer agent, wait for completion. The agent will:
- Read all research notes
- Synthesize findings
- Write comprehensive report to files/reports/
- Return completion message with file path
Phase 4: Deliver Results
Step 4.1: Create User Summary
# Research Complete: {Topic}
Comprehensive research completed with {N} specialized researchers.
## Key Findings
1. {Finding 1}
2. {Finding 2}
3. {Finding 3}
## Research Scope
{N} subtopics investigated:
- {Subtopic 1}
- {Subtopic 2}
- {Subtopic 3}
## Files Generated
**Research Notes**: `files/research_notes/`
- {file1}.md
- {file2}.md
- {file3}.md
**Final Report**: `files/reports/{filename}.md`
## Next Steps
{Optional suggestions}Step 4.2: Update TodoWrite Mark all items complete.
Best Practices
Good Decomposition
✅ 2-4 subtopics (sweet spot: 3) ✅ Distinct but related ✅ Comprehensive coverage ✅ Independently researchable
❌ Too many (>5) ❌ Too few (1) ❌ Significant overlap ❌ Too narrow or too broad
Parallel Execution
- Always spawn researchers simultaneously
- Never sequential unless dependent
- Provide context to each researcher
- Reasonable scope (10-15 min each)
Synthesis Quality
- Read ALL notes
- Find connections across subtopics
- Note contradictions explicitly
- Cite sources
- Add insights beyond individual notes
Error Handling
Researcher Fails: Try replacement, proceed with others, note gap No Results Found: Accept partial, note limitation Contradictory Findings: Document all perspectives explicitly Unclear Query: Ask clarifying questions first
Examples
Query: "Research quantum error correction" Decomposition:
- Theoretical foundations
- Hardware implementations
- Commercial viability Researchers: 3 parallel Synthesis: report-writer agent (MANDATORY)
Query: "Investigate cryptocurrency market 2025" Decomposition:
- Market metrics & players
- Regulatory landscape
- Technology evolution
- Institutional trends Researchers: 4 parallel Synthesis: report-writer agent (MANDATORY)
Remember: Quality depends on good decomposition, thorough researchers, insightful synthesis, and clear user communication.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

