BetterPrompt exhibits severe definition quality gaps across all 14 tools. Most tools lack visible input schemas in the provided source code, only 4 of 14 have documented parameters (scan-sessions, save-stage-output, get-stage-output, save-domain-results, save-user-prefs). Descriptions are present but generic and under-optimized for LLM selection (e.g., 'Extract structured data from parsed AI sessions' lacks WHEN/WHY context). No output schemas are documented. No error handling guidance, no parameter constraints (enums, ranges), no idempotent/destructive hints. Tool names follow verb_noun convention (positive), but composition issues exist: scan-sessions, extract-data, and get-prompt-context appear to form a pipeline without clear separation of concerns. No evidence of pagination, rate limiting, or recovery guidance. Security: no visible secret injection patterns, no permission gates, no audit trail declarations.

• Document complete input and output schemas for all 14 tools. Use JSON Schema with explicit types, descriptions, and constraints (enums, min/max, patterns).
• Expand tool descriptions to 50 - 200 characters. Include WHAT the tool does, WHEN to use it (vs similar tools), and any prerequisites. Example: 'Scan local AI coding sessions (Claude Code, Cursor) to extract conversation history. Call this first to discover available sessions before analysis.'
• Add parameter descriptions for all inputs. For scan-sessions, describe includeSources/excludeSources as enums (e.g., 'Array of source types: ["claude-code", "cursor", "vscode"]'), and minFileSize/maxFileSize with ranges (e.g., 'Minimum file size in bytes, 0 - 1GB').
• Declare tool annotations: mark read-only tools with readOnlyHint=true; mark write tools (save-*, sync-to-team) with destructiveHint=true; mark idempotent operations explicitly.
• Add error handling guidance to each tool description. Example: 'If no sessions found, returns empty array. If file access denied, returns error with available directories.' Include recovery hints for common failures.