Strong schema definitions with Zod validation and structured output. All 7 tools have descriptions and input schemas. Tool names follow verb_noun pattern (gc_submit_task, gc_get_task_status, etc.). However, descriptions vary in quality, some lack actionable context (e.g., gc_list_agents has no description of when to call it), and parameter descriptions are minimal. Output schemas are defined but not all include chaining IDs needed for multi-step workflows. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint) despite clear read/write semantics.

• Add tool annotations to all tools: mark gc_submit_task and gc_write_soul_capsule with destructiveHint=true; mark read-only tools with readOnlyHint=true. This guides agent planning and prevents accidental misuse.
• Enhance descriptions with actionable context. For gc_list_agents, add: 'Call this before gc_submit_task to see which agents handle your request_type, then choose request_type intentionally.' For gc_get_task_status, add: 'Known issue: some task IDs return stub data (Kovach-Enterprises/ambient-access-layer#1). If results seem incomplete, retry after 30s or check gc_get_job_status.'
• Document parameter relationships and expected values. For gc_submit_task.metadata, specify: 'Optional keys: priority (low|medium|high), timeout_seconds (1-3600), retry_count (0-5). Example: {"priority": "high", "timeout_seconds": 300}'.
• Add recovery guidance to error responses. Instead of 'Error: task not found', return 'Task not found. Verify task_id from gc_submit_task output, or call gc_get_telemetry to check if the task was routed. If still missing, the task may have expired (default 24h).'
• Include chaining IDs in output schemas. gc_submit_task should return trace_id (or document how to derive it from task_id) so agents can immediately call gc_write_soul_capsule without a lookup step.