MemForge defines 7 tools with partial schema coverage. All tools have descriptions (10-200 chars, within baseline), but input schemas are incomplete: most parameters lack explicit type definitions in visible code. The search tool references a custom 'RankedRetrievalIntent' type without definition. Parameters like 'time_range', 'source_filter', and 'entities' are documented as objects/arrays but lack internal field schemas. No output schemas are documented. Error handling is absent, no recovery guidance, categorization, or actionable error messages visible. Tool composition is reasonable (search, get, create, retire, propose, push variants), but parameter descriptions are generic ('List of memory types to filter by' lacks enum values or constraints). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) detected. Naming follows verb_noun convention (search, get_memory, create_memory, retire_memory, propose_memory_correction, push_*), clear and consistent.

• Document complete input schemas for all parameters. Define 'RangedRetrievalIntent', 'time_range', 'source_filter' with explicit field types, constraints, and examples. Use JSON Schema $defs for reusable types.
• Add output schemas for all tools. Specify which fields are returned (e.g., search returns [{ memory_id, content, memory_type, confidence, created_at, ... }]), their types, and which are required for downstream tool chaining.
• Implement error handling with recovery guidance. Return structured errors: { code, message, retryable, suggestion }. E.g., 'memory_not_found: Try search() with a partial content query.'
• Add tool annotations: mark retire_memory and propose_memory_correction as destructiveHint=true; mark search and get_memory as readOnlyHint=true; mark create_memory with idempotencyKey support as idempotentHint=true.
• Convert prose constraints to formal enums and bounds. memory_type: enum=['fact', 'decision', 'convention', 'procedure']; confidence: type=number, minimum=0, maximum=1; top_k: type=integer, minimum=1, maximum=100.