PAL-MEM exposes 5 well-named tools with clear, domain-specific descriptions (avg 180 chars). All tools have explicit input schemas with Zod validation and type definitions. However, output schemas are undocumented, responses are returned as unstructured text via asText() helper, forcing LLMs to parse JSON strings. Error handling is minimal (generic asError wrapper). Parameter descriptions are solid but lack format constraints (e.g., ISO date format for 'since' param not formally specified). Tool composition is clean (each does one thing), but no pagination support for recall/search despite potentially large result sets.

• Document output schemas explicitly. Define return type for each tool (e.g., palace_recall returns {query: string, k: number, hits: Array<{id, abstract, score, type, wing, hall}>}). Return structured objects, not JSON strings.
• Add pagination to palace_recall and palace_search: accept 'offset' and 'limit' params (default limit=10, max=50), return 'total' count and 'has_more' flag.
• Enhance error responses with recovery guidance. Instead of asError('Not found'), return {isError: true, code: 'NOT_FOUND', message: '...', suggestion: 'Try palace_search() with broader query'}.
• Formalize date format constraints. Add pattern: '^\d{4}-\d{2}-\d{2}T' to 'since' param schema, or use z.string().datetime() in Zod.
• Remove deprecated 'includeLowTrust' parameter after one release cycle. Update all tool descriptions to reference 'includeQuarantine' only.