Three tools with critical definition gaps. devkit_start and devkit_advance have minimal descriptions (under 100 chars) lacking WHEN/WHY context. spawn_agent has better structure but all three lack documented output schemas. Input schemas are present but sparse: devkit_advance has only one param with no type info visible; spawn_agent has enums but descriptions are generic. No error handling guidance, no recovery paths. Tool names are verb-based but descriptions fail to explain prerequisites, side effects, or expected return values. This server prioritizes workflow orchestration over agent clarity.

• Add output schema to devkit_advance documenting returned step object: {step_type, step_id, command?, prompt?, result?, status}. Include example.
• Expand devkit_start description to 150-200 chars: explain it initiates deterministic workflows, when to use vs spawn_agent, what happens on error, and that LLM must call devkit_advance in a loop until complete.
• Document devkit_advance step_result format: is it free-form text, JSON, or structured? Add constraint: 'Plain text or JSON object describing step outcome. Max 2000 chars.'
• Add error classification to all three tools: 'On error, returns {error, retryable: bool, suggestion}. If retryable=true, call devkit_advance again. If false, report to user.'
• For spawn_agent, provide role_definition format guidance: 'Markdown-formatted role instructions. Include: responsibilities, constraints, required output format, and scope boundaries.'