Retroscope has 7 well-named tools with consistent verb_noun patterns (retroscope_*). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present with typed parameters and descriptions. However, output schemas are NOT documented, responses are free-text strings with no structured field definitions. Parameter descriptions are adequate but lack format constraints (e.g., 'since' accepts '24h, 3d, 7d, or ISO date' but no regex/enum). Error handling is absent, no guidance on what to do if indexing fails or queries return empty. Security is reasonable (read-only tools, one write tool clearly marked), but no permission gates or audit trail hints. Composition is sound, each tool has one responsibility, though some could batch operations.

• Document output schemas for all tools. Define structured response objects with typed fields (e.g., standup returns {sessions: int, events: int, summary: string}). This enables LLMs to extract and chain data.
• Add enum constraints to 'since' parameter: ['24h', '3d', '7d', 'custom'] with optional 'custom_date' param for ISO dates. Prevents invalid format hallucinations.
• Add enum constraints to 'focus' parameter in tips/cost_tips/improve. Define valid categories explicitly: ['prompting', 'skills', 'sessions', 'planning'] for tips; ['cache', 'sessions', 'skills', 'subagents'] for cost_tips.
• Implement error handling with recovery guidance. Return structured errors: {error: string, code: string, recovery_hint: string}. E.g., 'No sessions found in past 24h. Try expanding the time window with since=7d or call retroscope_status to check index health.'
• Add dry_run parameter to retroscope_reindex to prevent accidental index rebuilds. Return preview of files to be reindexed before committing.