30 tools with minimal schema visibility and inconsistent descriptions. Most tools lack explicit input parameter schemas in the provided source, only 3 tools (codex_handoff_create, task_get, artifact_get) show partial schema hints. Descriptions are present but generic (10 - 50 chars), falling below the 50 - 200 char baseline for LLM-optimized docs. No output schemas documented. Error handling and recovery guidance absent. Tool naming follows verb_noun convention (good), but parameter descriptions are missing entirely for most tools. This server prioritizes API routing over agent-friendly tool design.

• Add explicit JSON Schema input definitions for all 30 tools. At minimum: type, description, and enum constraints for each parameter. Example: task_get should declare task_id as {type: 'string', description: 'UUID of the task to retrieve', pattern: '^[a-f0-9-]{36}$'}.
• Expand all tool descriptions to 80 - 150 characters. Include WHAT (action), WHEN (use case), and WHAT IT RETURNS. Example: 'Retrieves a task by ID. Use after task_claim_next to check status. Returns task object with id, status, runner, and lease_id.'
• Document output schemas for all tools. For list_* tools, include pagination (limit, offset, total_count). For task_* tools, return task_id, status, lease_id, and runner so downstream tools can chain calls.
• Add error handling guidance to WRITE tools (codex_handoff_create, task_claim_next, approval_resolve, etc.). Example: 'If lease_ttl_seconds exceeds 3600, returns 400 with message "lease_ttl_seconds must be ≤ 3600". Retry with a smaller value.'
• Add parameter descriptions and constraints. For task_claim_next: lane (enum: 'fake-codex'|'real-lane-1'|...), lease_ttl_seconds (integer, 1 - 3600), read_only (boolean, default false). For search_files: pattern (regex string, max 256 chars).