Comind exposes 8 code-intelligence tools with READ_ONLY risk profiles. All tools have descriptions (10 - 80 chars, mostly adequate) and input schemas with typed parameters. However, schemas lack depth: no enums for constrained inputs (e.g., 'format' in search accepts 'md' or 'table' but is declared as free-form string), no documented output schemas, and no error recovery guidance. Parameter descriptions are minimal (e.g., 'Index directory to read' lacks format/default clarity). Tool names are verb-noun (explore, search, find, repos, stats, guide, flow, changed) but lack specificity, 'explore' and 'flow' are vague compared to 'get_symbol_context' or 'trace_call_flow'. No tool declares idempotency, pagination, or rate-limit behavior. Error handling is absent from visible code, no guidance on what happens when a symbol is not found or an index is corrupted.

• Add output schemas for all tools. Document returned fields (symbol_id, name, kind, location, dependents_count, etc.) with types and descriptions so LLMs can chain calls.
• Convert constrained inputs to enums. Change 'format' in search to enum: ['md', 'table']. Add enums for symbol kinds (function, class, variable, etc.) in stats and guide.
• Rename tools for clarity: 'explore' → 'get_symbol_context', 'flow' → 'trace_call_flow', 'changed' → 'list_changed_files'. Verb-noun names with specificity reduce LLM confusion.
• Add error recovery guidance to descriptions. E.g., 'If symbol not found, try search() with a partial name.' Include examples of valid index_dir formats (local path or s3://bucket/path).
• Document pagination and limits. Specify max results per tool (e.g., search returns top 20 matches) and whether pagination is supported. Add limit and offset parameters if needed.