Critical gaps across all dimensions. Tool definitions are inferred from names and brief descriptions only, no explicit schema definitions visible in source code. Descriptions are minimal (10-20 chars), far below the 50-200 char baseline for LLM-optimized tools. No parameter schemas, types, or constraints documented. No output schemas visible. Error handling and recovery guidance absent. Tools lack the structure needed for reliable agent composition.

• Add explicit JSON Schema input definitions for all 14 tools. Each parameter must have type (string, number, boolean, array, object), description (50-100 chars), and constraints (enum, pattern, min/max). Example: dts_create_upload should document required params like 'pipeline_id' (enum of available pipelines), 'file_path' (string, path format), 'metadata' (object with schema).
• Expand tool descriptions to 100-200 characters. Include WHAT the tool does, WHEN to use it (e.g., 'Call this after dts_list_pipelines to initiate a new upload'), and any prerequisites. Example: 'Create a new upload session for a DTS pipeline. Call dts_list_pipelines first to get available pipeline IDs. Returns upload_session_id for use with dts_complete_upload.'
• Document output schemas for all tools. For list tools (dts_list_artifacts, dts_list_jobs, dts_list_pipelines), specify: array of objects with fields (id, name, status, created_at, etc.), pagination support (limit, offset, total_count), and max result limit (e.g., 50 items). For get tools, document all returned fields and their types.
• Add error handling guidance. For each tool, document: (1) what errors can occur (not found, permission denied, invalid input), (2) what the LLM should do next (retry, call a discovery tool, ask user). Example for dts_delete_artifact: 'If artifact not found, call dts_list_artifacts to verify the ID. If permission denied, the user lacks write access to this artifact.'
• Implement confirmation/dry-run for destructive operations. dts_delete_artifact and dts_cancel_job should support a 'dry_run' boolean parameter (default false) that returns what would be deleted/cancelled without executing. This prevents accidental data loss when agents make mistakes.