Qirrel provides 5 tools with complete JSON schemas and descriptions. All tools have proper input schemas with type definitions and required fields. Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present and correct. However, descriptions are brief (10-20 chars for tool names), and parameter descriptions lack actionable constraints (e.g., no min/max for concurrency, no format guidance for configPath). Output schemas are not documented. Error handling is minimal, no recovery guidance or categorization. Tool names use verb_noun pattern correctly (parse_text, parse_batch, speech_clean), but qirrel.capabilities and qirrel.tool_help are meta-tools that don't follow the primary action pattern.

• Document output schemas for parse_text, parse_batch, and speech_clean. Specify field names, types, and nesting (e.g., { entities: { type: string, phones: string[], emails: string[] }, metadata: { ... } }).
• Add min/max constraints to numeric parameters: concurrency (1 - 16), and document configPath format (e.g., 'YAML file path, relative to cwd or absolute').
• Wrap tool implementations with structured error responses: { error: string, code: 'INVALID_INPUT'|'NOT_FOUND'|'TIMEOUT', suggestion: string }. Return actionable guidance, not raw TypeErrors.
• Rename qirrel.capabilities → describe_qirrel_capabilities and qirrel.tool_help → list_qirrel_tools to follow verb_noun pattern and clarify intent.
• Add batch size limit documentation: 'texts array must contain 1 - 1000 items; each text ≤ 100k chars. Exceeding limits returns 400 with guidance.'