Single tool 'veritas_research' has a reasonable name and description, but critical gaps in schema documentation and parameter descriptions. The input schema is visible with three parameters (query, mode, sessionId), but parameter descriptions are minimal or missing context. Mode enum is present but lacks explanation of what 'lite', 'medium', 'deep' actually do in terms of research scope, cost, or latency. SessionId is marked optional but its purpose and lifecycle are unexplained. No output schema is documented, LLMs cannot plan downstream actions or extract structured results. Error handling and recovery guidance are absent. The tool lacks idempotency guarantees, timeout specifications, and rate-limit guidance critical for agent reliability.

• Document the complete output schema with field names, types, and descriptions. Example: {results: [{title, url, snippet, source, confidence}], total_sources: number, research_depth_used: string, execution_time_ms: number}.
• Expand mode parameter description: 'lite: quick surface-level search (1-3 sources, ~2s); medium: balanced research (5-10 sources, ~5s); deep: exhaustive multi-source verification (15+ sources, ~15s). Choose based on query complexity and time budget.'
• Clarify sessionId: 'Optional. Reuse a prior session ID to continue multi-turn research on the same topic, maintaining context and avoiding duplicate source checks. Omit for new independent queries.'
• Add error handling section to description: 'Returns error if query is empty, mode is invalid, or research times out. If no sources found, returns empty results array with explanation. Rate limit: 10 calls/minute per session.'
• Specify idempotency and retry safety: 'Idempotent within a 5-minute window, repeated calls with same query+sessionId return cached results. Safe to retry on timeout or transient errors.'