AgentSkillsCN

research-council

无需全盘辩论,即可实现轻量级的并行研究编排。 孵化拥有不同查询的预言家智能体,收集各方见解,最终合成研究报告。

SKILL.md
--- frontmatter
name: research-council
description: |
  Lightweight parallel research orchestration without full council debate.
  Spawns oracle agents with different queries, collects findings, synthesizes report.
model_invoked: true
triggers:
  - "research council"
  - "parallel research"
  - "multi-source research"
  - "investigate from multiple angles"
  - "comprehensive research"

Research Council Skill

Purpose

Orchestrate parallel research across multiple sources without the full Architecture Council process. Used when you need comprehensive research from different angles but don't need architectural debate and synthesis.

When to Use

Use Research CouncilUse Architecture Council
Pure information gatheringArchitectural decisions
Multiple search queriesDesign tradeoff analysis
External standards researchSpecification writing
Quick multi-angle investigationADR creation
Technology evaluationSystem design

Council Composition

AgentQuery FocusMCP
Internal OracleCodebase patternsGrep, Glob, Read
Docs OracleEffect documentationdeepwiki
Standards OracleIndustry standardsexa, playwright
Community OracleBest practices, examplesexa

Protocol

Phase 1: Query Definition

Define parallel research queries:

code
1. Internal: "How does our codebase handle X?"
2. Docs: "What does Effect documentation say about X?"
3. Standards: "What industry standards apply to X?"
4. Community: "What are common patterns for X?"

Phase 2: Parallel Research

Spawn oracle agents in parallel:

  • Each oracle queries its domain
  • Writes findings to shared report
  • Signals completion

Phase 3: Synthesis

Combine findings into research report:

  • Cross-reference sources
  • Identify consensus
  • Flag contradictions
  • Summarize actionable insights

Report Format

markdown
# Research Report: <Topic>

**Generated**: YYYY-MM-DD
**Author**: Research Council

---

## Executive Summary

[Key findings in 3-5 bullet points]

---

## Internal Analysis

**Query**: "How does our codebase handle X?"

### Findings

[Codebase patterns found]

### Codebase References

| File | Relevance |
|------|-----------|
| `src/lib/...` | Description |

---

## Documentation Research

**Query**: "What does Effect documentation say about X?"
**Source**: deepwiki: Effect-TS/effect

### Findings

[Documentation patterns]

### Verification Status

- [x] API verified current
- [ ] Breaking changes checked

---

## Industry Standards

**Query**: "What industry standards apply to X?"
**Source**: exa search, ISA/FDA/ISO references

### Findings

[Standards references]

### Applicable Standards

| Standard | Relevance | Link |
|----------|-----------|------|
| ISA-95 | ... | ... |

---

## Community Patterns

**Query**: "What are common patterns for X?"
**Source**: exa search, GitHub, blog posts

### Findings

[Community patterns]

### References

1. [Reference 1]
2. [Reference 2]

---

## Cross-Reference Analysis

### Consensus

[What all sources agree on]

### Contradictions

[Where sources disagree]

### Gaps

[What no source addressed]

---

## Recommendations

1. [Actionable recommendation]
2. [Actionable recommendation]

---

## Sources

[Full citation list]

Example Invocation

User Request

"Research how event sourcing is used in IIoT manufacturing systems"

Claude Response

code
I'll convene a Research Council to investigate this from multiple angles.

**Research Queries**:
1. Internal: "Event patterns in our IIoT codebase"
2. Docs: "Effect EventLog patterns"
3. Standards: "ISA-95 event sourcing manufacturing"
4. Community: "IIoT event sourcing best practices"

[Oracles run in parallel]

**Report generated**: thoughts/shared/reports/2026-01-26-es-iiot-research.md

MCP Tool Usage

deepwiki (Documentation)

code
mcp__deepwiki__ask_question
  repoName: "Effect-TS/effect"
  question: "What is the recommended pattern for EventLog in production use?"

exa (Web Search)

code
mcp__exa__search
  query: "ISA-95 event sourcing manufacturing operations management 2024"
code
mcp__exa__search
  query: "event sourcing IIoT best practices industrial"

playwright (Standards Scraping)

For regulatory documents that need scraping:

code
mcp__playwright__navigate
  url: "https://www.isa.org/standards/..."

Query Templates

Technology Evaluation

code
Internal: "Do we already use {technology}?"
Docs: "What is {technology} designed for?"
Standards: "{technology} in {domain} standards"
Community: "{technology} production experience reviews"

Pattern Investigation

code
Internal: "How do we implement {pattern}?"
Docs: "Recommended implementation of {pattern}"
Standards: "Industry standards for {pattern}"
Community: "{pattern} best practices {domain}"

Problem Research

code
Internal: "Do we have this {problem}?"
Docs: "Known issues with {related-feature}"
Standards: "Industry solutions for {problem}"
Community: "{problem} solutions {technology}"

Lightweight vs Full Council

Research Council (This Skill)

code
Queries → Oracles → Report
  • ~10-15 minutes
  • Information gathering
  • Report output

Architecture Council (Full)

code
Documents → Agents → Journal → Synthesis → Spec → ADR → WBS
  • ~30-60 minutes
  • Decision making
  • Multiple artifacts

Output Location

Reports go to:

code
thoughts/shared/reports/YYYY-MM-DD-<topic>.md

Related Skills

SkillIntegration
/architecture-councilUpgrade to full council when decisions needed
/grounded-researchSingle-source research
/effect-researchEffect-specific research

Anti-Patterns

DON'T: Use for decisions

code
❌ "Should we use X or Y?" → Use Architecture Council
✅ "What is X and Y?" → Use Research Council

DON'T: Skip internal analysis

code
❌ Only search externally
✅ Always check codebase first

DON'T: Ignore contradictions

code
❌ "Source A says X"
✅ "Source A says X, but Source B says Y. Resolution needed."