14 tools with mixed quality. Naming is generally verb-based (score_item, enrich_concept, wiki_search) but some are vague (trigger_codegarden_drift, dsh_session). Descriptions exist for all tools but many are technical jargon-heavy and lack LLM-friendly guidance on WHEN to use them. Input schemas are present with types and constraints (enums, min/max), but output schemas are undocumented. Error handling is minimal, most tools return generic 500 errors without recovery guidance. Security: path traversal mitigations visible in enrich_concept and wiki_read, but no audit logging or permission gates documented.

• Document output schemas for all 14 tools. For each, specify returned fields, types, and whether results are paginated. Example: 'Returns {status: string, score_id: string, created_at: ISO8601}.'
• Rewrite descriptions in English, 50 - 200 characters, answering: What does it do? When should the LLM call it? What does it return? Example: 'Search knowledge items by keyword using full-text search. Returns up to 20 matching items with titles and IDs. Use before wiki_read to find relevant content.'
• Rename trigger_codegarden_drift to evaluate_project_tech_drift and dsh_session to get_dsh_session_status. Use verb_noun pattern consistently.
• Add structured error responses with recovery guidance. Instead of HTTPException(500, 'write_score failed'), return {error: 'hotspot_id not found', suggestion: 'Call search_hotspots() first', retryable: false}.
• Add permission checks and audit logging. Document required scopes (e.g., 'Requires write:knowledge scope') and log caller, timestamp, parameters, and outcome for all write tools.