Three tools with complete schemas and descriptions, but significant gaps in parameter documentation and error guidance. web_task_claim and web_task_list have no parameters (empty objects), limiting composability. web_task_settle has well-defined parameters with descriptions and enum constraints for stopReason. Descriptions are adequate (100-200 chars) but lack dependency hints and recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Error handling exists but is minimal, resolveTaskId returns error strings rather than structured guidance. Output schemas are documented as strings. Tools follow a clear claim/settle pattern but lack idempotency guarantees and batch operations.

• Add optional parameters to web_task_claim: 'timeout_ms' (default 30000, range 1000 - 300000) to let agents control wait duration, and 'filter_label' to claim tasks matching a label pattern.
• Document web_task_list output schema explicitly: { type: 'array', items: { type: 'object', properties: { id, label, cwd, prompt, status }, required: ['id', 'status'] } }. Add 'limit' (1 - 100, default 20) and 'offset' (default 0) parameters for pagination.
• Replace string error responses with structured objects: { error: string, code: 'no_task_available'|'ambiguous_task_id'|'task_not_found', recovery: string }. E.g., 'recovery': 'Call web_task_claim again to wait for work.'
• Add tool annotations: web_task_claim and web_task_list get readOnlyHint=true; web_task_settle gets destructiveHint=true and idempotentHint=false (settling twice fails, not idempotent).
• Add dependency hints to descriptions: 'Call web_task_list first to see queued tasks and their IDs before claiming.' Include example task ID format ('task-abc123') in web_task_settle description.

Inferred effective spec: 2025-06-18+.

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained