Research Orchestration

Execute parallel research agents to gather comprehensive insights on a topic, then synthesize into actionable reports.

Shared Orchestration Patterns

This skill builds on shared orchestration patterns. See:

• includes/orchestration/parallel-execution.md - Agent dispatch and concurrency
• includes/orchestration/result-synthesis.md - Collecting and merging results
• includes/orchestration/context-management.md - Context folding and memory
• includes/orchestration/checkpoint-patterns.md - Save/resume workflows

The patterns below are research-specific extensions.

Research-Specific Workflow

Phase 1: Topic Analysis

1. Parse user query for research scope
2. Determine which agents to dispatch:
   • Always: technical, architectural
   • If UI/UX involved: ux
   • If security implications: security
   • If strategic/market question: competitive

Phase 2: Parallel Research

Dispatch selected agents with:

• Topic context
• Current year requirement (2026)
• Confidence scoring requirement
• Citation requirement

Phase 3: Synthesis

After all agents complete:

1. Launch research:synthesis agent
2. Cross-validate findings
3. Generate consolidated report
4. Write to docs/.claude/research/{topic}/

Agent Selection Matrix

Query Contains	Agents to Dispatch
"how to implement"	technical, architectural
"best practice"	technical, security
"UI", "UX", "user"	technical, ux
"secure", "auth"	technical, security
"compare", "vs"	technical, competitive
"market", "trend"	competitive
Default	technical, architectural

Output Structure

docs/.claude/research/{topic-slug}/
├── README.md          # Executive summary
├── technical.md       # Technical findings
├── architectural.md   # Architecture analysis
├── ux.md             # UX considerations (if applicable)
├── security.md       # Security analysis (if applicable)
├── competitive.md    # Market context (if applicable)
└── sources.md        # All citations

Critical Requirements

Year Validation

ALL web searches MUST include current year:

WebSearch(query: "{topic} best practices 2026")

Confidence Scoring

Every finding must include:

• HIGH: Multiple 2025-2026 sources agree
• MEDIUM: Some sources, needs verification
• LOW: Single source or older

Citations

Every finding must cite:

• Source URL
• Access date
• Relevance to query

Integration Points

Before Brainstorm

/brainstorm "add feature X"
  -> Automatically triggers /research "feature X" first
  -> Brainstorm receives research context

During Troubleshooting

Error encountered
  -> /research:debug "error message"
  -> Technical + architectural agents analyze

Command Variants

Command	Agents	Use Case
/research "topic"	Auto-selected	General research
/research:quick "topic"	technical only	Fast answer
/research --agents tech,security "topic"	Specified	Targeted
/research:debug "error"	tech, arch	Troubleshooting

Consensus Review (Required)

After synthesis completes, run consensus review:

Review Agents

• documentation-reviewer - Technical accuracy, completeness
• code-quality-reviewer - Code example correctness (if applicable)

Pass Criteria

Requirement	Threshold
Consensus Score	>= 70%
Critical Findings	0
High Findings	0

Re-Synthesis Loop

If consensus < 70% OR critical/high findings exist:

1. Collect reviewer feedback
2. Re-run synthesis agent with feedback
3. Re-run consensus review
4. Maximum 3 iterations
5. Report partial success if max reached

See includes/orchestration/consensus-review.md for detailed patterns.

Anti-Patterns

• Searching without current year -> outdated results
• Skipping synthesis -> disconnected findings
• Not checking codebase -> recommendations don't fit
• Single source confidence HIGH -> inaccurate scoring
• Skipping consensus review -> quality not validated