claude-vault has 6 tools with significant definition gaps. Tool names lack clear action verbs (sessions_ingest, sessions_query, sessions_stats, sessions_report, sessions_tag, sessions_filter use prefix pattern but lack verb clarity). Descriptions exist but are brief (avg ~120 chars) and lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Parameter descriptions are sparse: sessions_query has 'sql' and 'limit' described, but sessions_stats has only optional 'project' and 'since' with minimal context. No output schemas documented anywhere. Error handling is absent, no guidance on what happens on SQL errors, permission failures, or invalid filters. No tool annotations (readOnlyHint/destructiveHint) despite clear READ vs WRITE semantics.

• Rename tools to verb-first: ingest_sessions, query_sessions, get_session_stats, generate_session_report, tag_sessions, filter_sessions_by_tag. This matches the 90% baseline of A+ tools starting with action verbs.
• Add explicit input schemas to sessions_ingest and sessions_tag. Even if they take no parameters, use {"type":"object","properties":{},"required":[]} instead of {}.
• Document output schemas for all tools. E.g., sessions_query returns {"rows": [{"session_id": string, "event_type": string, ...}], "total": number}. sessions_stats returns {"total_tokens": number, "estimated_cost": number, "duration_seconds": number, "session_count": number}.
• Add tool annotations: mark sessions_ingest and sessions_tag with destructiveHint=true (they modify state); mark others with readOnlyHint=true. This enables agents to reason about safety.
• Expand parameter descriptions. sessions_stats 'project' should say 'Filter by project slug (optional; if omitted, aggregates all projects)'. sessions_query 'limit' should say 'Max results (default 100, max 1000; larger queries may timeout)'.