Critical gaps across all 19 tools. All descriptions are identical boilerplate ('Production Master investigation tool: <name>'), providing zero context for LLM tool selection. Input schemas present but severely incomplete: most WRITE tools (add_evidence, add_correction, add_comment, create_snapshot, fork, etc.) declare only investigationId as required, omitting all actual payload parameters (content, data, etc.). Parameters lack descriptions entirely. No output schemas documented. No error handling guidance. Naming is acceptable (verb_noun pattern), but descriptions fail the 20-character minimum for actionability.

• Replace all boilerplate descriptions with specific, actionable text (50 - 200 chars). Example: 'Get a summary of the investigation including status, key findings, and timeline. Call this first to understand the investigation scope.' vs 'Production Master investigation tool: get_summary'.
• For WRITE tools, expand input schemas to include all required and optional payload fields. Example: add_evidence should include evidenceType, content, sourceUrl, confidence, etc. with type definitions and descriptions.
• Add descriptions to every parameter. Example: 'investigationId (string): The unique identifier of the investigation. Obtain from list_investigations or from a previous tool response.'
• Document output schemas for all tools. Example: 'Returns {investigationId, status, createdAt, updatedAt, summary, keyFindings, timeline}' so LLMs know what fields to expect.
• Add error handling guidance to WRITE tools. Example: 'On 404: investigation not found, call list_investigations to find the correct ID. On 409: conflict, the investigation may have been modified; call get_summary to refresh state.'