Primer exposes 5 tools with inconsistent quality. Tool names lack clear action verbs (seed_now, reset_agents are vague; session_pending is passive). Descriptions vary widely: seed_now and reset_agents have detailed, context-rich descriptions (100+ chars), but session_pending is terse (34 chars). Input schemas are present but sparse: seed_now and reset_agents have empty objects ({}), while list_external_tool_calls has 5 parameters with type info but no parameter descriptions visible. No output schemas documented. Error handling and recovery guidance absent. Composition is weak: reset_agents modifies two rows with side effects but no idempotency guarantee stated; seed_now's 'S5 ensure pass' is domain-specific jargon not explained. No security annotations (permissions, scope declarations). Overall, definitions are functional but fall short of production-grade LLM-optimized tooling.

• Rename tools with clear action verbs: seed_now → ensure_seeded_world, reset_agents → reset_agent_definitions, session_pending → list_pending_tool_calls, setup_predicates → check_setup_predicates.
• Document output schemas for all tools. E.g., seed_now should return {status: string, setup_state: object}; list_external_tool_calls should return {calls: [{id, status, session_id, chat_id, ...}], total: int, next_offset: int}.
• Add parameter descriptions to list_external_tool_calls: explain valid status values (pending, completed, failed?), clarify offset/limit constraints (max 200), and document filter semantics (AND vs OR logic).
• Add error handling guidance: 'If setup_predicates returns degraded=true, the LLM provider is unreachable. Retry after 30s or ask the user to check connectivity.' Include recovery paths for each tool.
• Declare tool permissions and idempotency: reset_agents should state 'Idempotent: safe to retry. Requires admin scope.' seed_now should clarify whether repeated calls are safe or have side effects.