68 tools with highly inconsistent quality. Only 4 tools (get_me, enable_toolset, list_available_toolsets, get_toolset_tools) have visible descriptions and input schemas in source. Remaining 64 tools lack visible schema definitions and descriptions in provided code, only tool names and risk classifications are shown. This creates a floor of ~6% per tool (naming only). Average across all 68 tools: (4 tools × ~65) + (64 tools × ~6) / 68 ≈ 42. No evidence of parameter descriptions, output schemas, or error recovery guidance for the majority.

• Add full JSON Schema input definitions to all 64 tools in pkg/github/tools.go. Include type, description, required fields, and constraints (enums, min/max, patterns) for every parameter.
• Write 50 - 200 character descriptions for every tool explaining WHAT it does, WHEN to use it, and any prerequisites. Example: 'Search GitHub repositories by name, language, stars, or date. Returns up to 30 results; use pagination for more. Requires read:repo scope.'
• Add pagination support (limit, offset/cursor, total_count) to all list_* and search_* tools. Document max result limits (e.g., 'Returns up to 30 results per page') in descriptions.
• Split 'create_and_submit_pull_request_review' into 'create_pull_request_review' + 'submit_pull_request_review' to follow single-responsibility principle. Rename 'add_pull_request_review_comment_to_pending_review' to 'add_review_comment' for clarity.
• Add error recovery guidance to all tools. Example: 'If repository not found, call search_repositories() with partial name. If permission denied, verify GitHub token has repo scope.' Include actionable error messages with valid options.

Inferred effective spec: <=2025-11-25.

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