agent-clis is a STDIO-only MCP server with 10 tools for task/intent coordination. Tool definitions are present with schemas and descriptions, but quality is inconsistent. Most tools have adequate descriptions (50-150 chars), but parameter descriptions are sparse or missing context. Schemas are properly typed but lack validation constraints (enums, ranges). No error handling guidance, no output schema documentation, and no tool annotations. The assign/claim tools handle complex state (conflict policies, path globs) but lack actionable error recovery patterns. Overall, definitions are functional but below production baseline.

• Document output schemas for all tools. E.g., 'assign' returns {assignment_id, task_id, agent_id, status, created_at}. LLMs need this to plan downstream calls.
• Add error recovery guidance. E.g., 'If conflict detected, call list-conflicts to see active claims, then retry with supersede parameter.'
• Add validation constraints to parameter descriptions. E.g., 'reason: string (max 500 chars, privacy-safe, no PII)' and 'paths: array of glob patterns (e.g., /home/user/*, /tmp/**).'
• Implement pagination for list tools. Add 'limit' (1-100, default 20) and 'offset' (default 0) parameters; return 'total_count' and 'has_more' in response.
• Add tool annotations via MCP schema. Mark 'verify', 'list-*' as readOnlyHint=true; mark 'assign', 'claim', 'close' as destructiveHint=true; mark 'heartbeat' as idempotentHint=true.