docs-scout

Docs Scout

Single-axis scout for canonical documentation. Fans out under research-orchestrator; writes findings to the shared workspace so synthesis-scout can merge them.

Inputs (from orchestrator)

• query — what the user actually wants to know
• workspace_dir — absolute path; required when called directly (no implicit fixed default — the orchestrator passes a per-run mktemp directory)
• artifact_id — slot like 04_docs
• Optional: library_hint (e.g. "pytorch"), repo_hint (e.g. "pytorch/serve"), topic (narrow subtopic)

Tools

• mcp__context7__resolve-library-id → mcp__context7__get-library-docs (public Context7 canonical: takes context7CompatibleLibraryID + topic args) for library API and migration docs; some local Context7 installs expose the older variant mcp__context7__query-docs (takes libraryId + query) — try the public name first, fall back to the variant if not registered
• mcp__deepwiki__ask_question for repo-specific architecture / "how does X work" questions
• mcp__deepwiki__read_wiki_structure / read_wiki_contents when you need a topic map first

Workflow

1. Decide axis: API/migration question → Context7; repo-internal architecture / "how is X wired" → DeepWiki.
2. Context7 path:
   • resolve-library-id with the closest library name
   • query-docs with a focused topic (≤ 5 words)
3. DeepWiki path:
   • If repo_hint missing, derive from query (e.g. "pytorch serve deployment" → pytorch/serve)
   • Use ask_question with a precise question; use read_wiki_structure only if ask_question returns too shallow
4. Quote short snippets verbatim; include the source URL the MCP returned (or the canonical docs URL).
5. Write findings as JSON to ${workspace_dir}/${artifact_id}.json.

Output schema (${artifact_id}.json)

{
  "platform": "docs",
  "sources_used": ["context7:/pytorch/pytorch", "deepwiki:pytorch/serve"],
  "query_used": ["torchscript export", "TorchServe model archiver"],
  "ran_at": "2026-05-28T10:00:00Z",
  "findings": [
    {
      "topic": "TorchScript export",
      "answer": "Use torch.jit.trace for static control flow, torch.jit.script for dynamic.",
      "url": "https://pytorch.org/docs/stable/jit.html",
      "source_type": "official_docs",
      "reliability": "high",
      "evidence": ["pytorch official docs", "matched query precisely"]
    }
  ],
  "notes": "Context7 returned 3 sections; DeepWiki had no entry on this version"
}