external-context

Fetch external documentation, references, and context for a query. Decomposes the query into 2-5 independent facets and spawns parallel document-specialist agents that search the web and synthesize findings.

<Use_When>

• User says "external-context", "look up", "research docs", "search the web"
• A question requires official documentation, not just codebase context
• Multiple independent aspects of a topic need research simultaneously
• Verification of facts against official sources is needed </Use_When>

<Do_Not_Use_When>

• The answer is in the codebase (use explore instead)
• The question is trivial (just answer it directly)
• Only one aspect of the topic needs research (parallelism adds overhead)
• User wants opinions rather than factual documentation </Do_Not_Use_When>

<Why_This_Exists> Codebase context is limited. External documentation, official specs, and web research provide the full picture. Parallel research is faster than sequential research. </Why_This_Exists>

<Execution_Policy>

• Decompose into 2-5 independent facets for parallel research
• Spawn document-specialist agents in parallel (max 5)
• All findings must cite sources with URLs
• Synthesize results into a coherent answer </Execution_Policy>

Phase 1: Facet Decomposition

Given a query, decompose into 2-5 independent search facets:

## Search Decomposition

**Query:** <original query>

### Facet 1: <facet-name>
- **Search focus:** What to search for
- **Sources:** Official docs, GitHub, blogs, etc.

Phase 2: Parallel Agent Invocation

Fire independent facets in parallel via Agent:

Agent(subagent_type="oh-my-claudecode:document-specialist", model="sonnet", prompt="Search for: <facet 1 description>. Use WebSearch and WebFetch to find official documentation and examples. Cite all sources with URLs.")

Maximum 5 parallel document-specialist agents.

Phase 3: Synthesis Output Format

Present synthesized results:

## External Context: <query>

### Key Findings
1. **<finding>** — Source: [title](url)
2. **<finding>** — Source: [title](url)

### Detailed Results

#### Facet 1: <name>
<aggregated findings with citations>

#### Facet 2: <name>
<aggregated findings with citations>

### Sources
- [Source 1](url)
- [Source 2](url)

<Tool_Usage>

• Use Agent with oh-my-claudecode:document-specialist for parallel web research
• Use WebSearch for initial search queries
• Use WebFetch to retrieve specific pages
• Maximum 5 parallel agents to avoid rate limits </Tool_Usage>

``` /oma:external-context What are the best practices for JWT token rotation in Node.js? /oma:external-context Compare Prisma vs Drizzle ORM for PostgreSQL /oma:external-context Latest React Server Components patterns and conventions ```

<Final_Checklist>

• Query decomposed into 2-5 facets
• All facets fired in parallel
• All findings cite sources with URLs
• Results synthesized into a coherent answer
• No facet is skipped due to parallelism limits </Final_Checklist>