Three tools with complete schemas and descriptions, but quality varies. start_process and kill_process have detailed, actionable descriptions (120-180 chars) with comprehensive parameter schemas including enums, ranges, and type info. read_output similarly well-documented. However, parameter descriptions are sometimes generic ('Process identifier from start_process' lacks context on format/lifetime). No output schemas documented, LLMs cannot predict response structure for chaining. Error handling guidance absent; no recovery hints. Naming is clear (verb_noun pattern) but 'start_process' is slightly generic vs domain-specific alternatives. No tool annotations (readOnlyHint, destructiveHint) despite clear risk profiles (IRREVERSIBLE, READ_ONLY marked in metadata but not in schema).

• Document output schemas for all three tools. Example: start_process returns {process_id: string, status: 'running'|'completed', exit_code?: number, stdout?: string, stderr?: string}. This enables LLMs to chain calls and extract required fields.
• Add error recovery guidance to descriptions. E.g., kill_process: 'Terminate a running process. If process_id is invalid, returns error with available process IDs. Retry with correct ID or call start_process to launch a new process.'
• Expose tool annotations in schema. Add 'destructiveHint: true' to start_process and kill_process; 'readOnlyHint: true' to read_output. This signals safety constraints to LLM clients without requiring description parsing.
• Clarify vague parameter descriptions. Replace 'Activity tracking target' with 'Optional object for tracking execution context (e.g., {user_id: string, session_id: string}). Used for audit and correlation.' Replace 'Action classification' with 'Optional string tag for categorizing the action (e.g., "user-initiated", "scheduled"). Used for logging and filtering.'
• Add idempotency guidance. Document which tools are idempotent (read_output) vs non-idempotent (start_process, kill_process). For non-idempotent tools, suggest deduplication strategies (e.g., 'Use process_id to deduplicate retries; calling start_process twice with identical args may launch two processes').