The server exposes a single tool 'assign_task' with major quality gaps. The tool definition is visible in client.py but lacks critical elements: the description (82 chars) is present but vague and lacks actionable guidance for LLM selection; parameters have minimal description ('The task description' and 'The desired format'); no output schema is documented; no error handling or recovery guidance; no input validation or constraints (e.g., task length limits, format enum); the implementation is entirely mock/local and does not actually execute tasks. The tool name 'assign_task' is verb-noun styled correctly, but the description does not explain WHEN to use this vs other task-handling patterns, what the actual execution plan represents, or how deliverable_format influences behavior. No examples of valid format values provided (should be enum). The parameters lack type constraints and detailed descriptions per the 72-char baseline. Output is undocumented, the response includes 'plan', 'estimated_duration_min', and 'coworker_id', but there is no specification of these fields' types, semantics, or guarantees. The mock implementation generates trivial placeholder data (parsing task by word count, fabricating step labels) with no actual task execution capability.

• Add a detailed output schema documenting 'plan' (array of strings), 'estimated_duration_min' (integer, range 5 - 1440), 'coworker_id' (string, format: 'clark-{8-hex}'), and 'status' (enum: 'queued', 'running', 'completed'). Include in tool registration.
• Convert 'deliverable_format' to an enum parameter: ['Markdown Report', 'JSON', 'PDF', 'Email', 'Confluence Page']. Add description: 'The format for the final deliverable. JSON is machine-readable; Markdown is human-friendly. Choose based on downstream consumption.'
• Expand tool description to 150 - 200 chars: 'Assigns a task to the Clark AI coworker for execution. Use when you need sustained work (research, analysis, reporting) rather than quick lookups. Returns an execution plan and estimated duration. Monitor status with get_task_status or cancel with cancel_task.'
• Expand parameter descriptions: task → 'A detailed task description (max 500 chars). Can include steps, constraints, or expected output. E.g., "Research top 10 AI frameworks and compare on cost, latency, and ease of integration."' deliverable_format → 'The desired output format. Use Markdown Report for summaries, JSON for structured data, PDF for shareable documents.'
• Implement actual task execution or connect to a real backend service. Mock implementations cannot be trusted by agents. If this is a demo, document that in the server README and mark tools as unsafe for production.
• Add error responses with actionable guidance: 'Task too large (501+ chars). Break into smaller subtasks or request a batch assignment.' 'Invalid format: must be one of [list]. Did you mean Markdown Report?'