| name | scout |
| description | Isolated research agent for external sources (Exa + Context7). |
| agent | .claude/agents/scout.md |
Scout Skill (Wrapper)
Invokes scout subagent for isolated research via Exa + Context7.
Architecture: This skill is a WRAPPER over .claude/agents/scout.md.
The agent file is the source of truth for the scout prompt.
When to Use
During Spark: Before designing feature, gather external knowledge
During Development: When need library docs or patterns
Invocation
Task tool:
description: "Scout research: {topic}"
subagent_type: "scout"
prompt: |
MODE: quick | deep
QUERY: {research question}
TYPE: library | pattern | architecture | error | company | general
DATE: {current date}
Research Tools
Scout has access to:
Exa (web research):
mcp__exa__web_search_exa — web search with clean content
mcp__exa__web_search_advanced_exa — filtered search (date, domain)
mcp__exa__get_code_context_exa — code from GitHub, StackOverflow
mcp__exa__deep_search_exa — deep search with query expansion
mcp__exa__crawling_exa — extract content from specific URL
mcp__exa__company_research_exa — company information
mcp__exa__deep_researcher_start — start AI researcher (complex topics)
mcp__exa__deep_researcher_check — get deep research results
Context7 (library docs):
mcp__plugin_context7_context7__resolve-library-id — find library ID
mcp__plugin_context7_context7__query-docs — official documentation
Local:
- Read, Glob, Grep — codebase exploration
Output
{
"tldr": "One sentence: the best solution/answer",
"recommendation": {
"solution": "Recommended approach",
"why": "2-3 sentences rationale",
"confidence": "high | medium | low",
"caveats": ["Limitation 1"]
},
"alternatives": [
{"name": "Alt", "pros": [], "cons": [], "when_to_use": "..."}
],
"sources": [
{"title": "...", "url": "...", "type": "docs|blog|official", "relevance": "..."}
],
"triangulation": {
"verified_claims": ["Claim (confirmed by N sources)"],
"conflicting_info": ["Topic: source1 vs source2"]
}
}
Notes
- Scout isolates web search "garbage" from main context
- Use quick mode for simple questions, deep for complex topics
- Always returns structured output with sources