Manicule has 5 tools with basic verb-noun naming (collection_list, search, ask, document_create). All tools have descriptions (10-100 chars), but most are terse and lack actionable context for LLM selection. Input schemas are present for 4/5 tools but lack parameter descriptions entirely, the 'search' tool schema is inferred from pyproject.toml rather than visible in source code. Output schemas are undocumented. No error handling guidance, no pagination support despite search returning potentially large result sets, and no indication of which tools are idempotent or safe to retry. The 'ask' tool conflates search + synthesis into one operation, violating single-responsibility. Tool composition is weak: no chaining IDs documented, no recovery paths for failures.

• Add 50-150 character descriptions to every parameter. Example: 'query (string): The search query, natural language or keywords. Required.' instead of omitting descriptions entirely.
• Document output schemas for all tools. Specify field names, types, and whether they are required. Example: 'Returns {results: [{id, title, score, collection}], total_count, next_cursor}'.
• Split 'ask' into 'search' (retrieval only) and 'generate_answer' (synthesis from search results). This enables agents to compose and retry independently.
• Add pagination: accept 'offset' and 'limit' for search(), return 'total_count' and 'next_cursor'. Cap default limit at 20-50 results to prevent context explosion.
• Add error handling guidance to tool descriptions. Example: 'If collection not found, call collection_list() to see available collections. If search returns no results, try broader keywords or different collection.'