This bash-based STDIO server has severe definition quality gaps. Tool descriptions are present but minimal (10-50 chars, well below the 50-200 char LLM-optimized baseline). Input schemas exist for most tools but lack depth: many parameters have no type constraints, enums are missing where appropriate (e.g., 'format' in report tool), and descriptions are generic. The 'init' tool accepts 'category' and 'tag' as free-form strings with no enum or validation guidance. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Output schemas are undocumented. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. The 'reset' and 'abort' tools are destructive but lack confirmation patterns or dry-run support.

• Expand tool descriptions to 50-200 chars. Example: 'next' → 'Retrieve and return the next test case to execute. Call this after init() to begin test sequence. Returns test name, action, and expected result.'
• Add output schema documentation for each tool. Specify returned fields, types, and whether results are paginated. Example: 'status returns {passed: int, failed: int, total: int, current_test: string}'.
• Add tool annotations: mark 'reset' and 'abort' with destructiveHint=true; mark 'next', 'status', 'vars' with readOnlyHint=true; mark 'validate' with idempotentHint=true.
• Implement confirmation pattern for destructive tools. Add optional 'confirm' boolean parameter to 'reset' and 'abort', or return a confirmation token requiring explicit re-invocation.
• Add error handling guidance. When tools fail, return structured errors: {error: string, code: string, retryable: bool, suggestion: string}. Example: 'No more tests. Call init() to start a new run.'