RegAssist exposes 3 tools with basic descriptions and parameter annotations, but critical gaps undermine production readiness. All tools have descriptions (good), but they lack output schema documentation, error handling guidance, and actionable recovery paths. Parameter descriptions are present but generic. Two of three tools are stubs with TODO comments indicating incomplete implementation. The search tool lacks pagination/result-limit documentation despite returning ranked chunks. fetch_document and summarize lack clear output schemas. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.

• Document output schemas for all three tools. E.g., search returns {chunkId, act, structuralPath, sourceUrl, score, text}[]; fetch_document returns {documentId, act, structuralPath, sourceUrl, text}; summarize returns string. Include these in tool descriptions or a separate schema reference.
• Complete stub implementations. fetch_document and summarize currently return TODO placeholders. Wire them to IChatService and the document store as indicated in comments.
• Add pagination to search: accept limit (default 20, max 100) and offset parameters; return total count and next_offset in response. Document in description: 'Returns up to 20 ranked chunks per call.'
• Add error recovery guidance. E.g., search: 'If no results, try broader keywords or omit actFilter.' fetch_document: 'If documentId not found, call search() first to get valid IDs.'
• Annotate tools with readOnlyHint=true in tool registration to signal they are safe for agents to call without confirmation.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry