Mecha provides 19 tools with basic schemas and descriptions, but quality is inconsistent. Most tools have descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization) and input schemas with type definitions. However, parameter descriptions are sparse or missing entirely. For example, kg_search has 8 parameters but only 3-4 have meaningful descriptions; 'probe' and 'include_private' lack context on when to use them. Output schemas are not documented anywhere in the source. Error handling is minimal, no recovery guidance, no categorization of retryable vs fatal errors. Tool composition is reasonable (separate read/write/delete operations), but naming could be clearer: 'kg_*' prefix is domain-specific but doesn't follow verb_noun convention (should be 'search_kg', 'list_kg_tasks'). Shell and fs_* tools expose destructive operations without confirmation patterns. No evidence of permission gates or audit logging.

• Add 50-200 character descriptions to all tools explaining WHAT they do, WHEN to use them, and any prerequisites. Current descriptions (10-50 chars) are too terse for LLM selection.
• Document output schemas for all tools. Specify field names, types, and what data is returned. This enables agents to plan downstream calls and extract the right data.
• Add parameter descriptions to all parameters. For kg_search, explain 'probe' (whether to fetch source metadata), 'include_private' (whether to include private sources), and 'scope' (evidence_only vs all entities).
• Implement confirmation or dry-run patterns for destructive tools (shell, fs_write, fs_edit, calendar_delete_event, mail_triage). Return a confirmation request before executing irreversible operations.
• Rename tools to follow verb_noun convention: kg_task_list → list_kg_tasks, kg_search → search_kg, mail_reply → reply_to_mail. This helps LLMs infer intent from the name alone.

Inferred effective spec: 2026-07-28+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry