Three tools with severely deficient schemas and descriptions. gdrive_connector and GPT5-gdrive-fixer have vague, legacy-focused descriptions (under 200 chars but lacking actionable context). GPT5-web-search is minimal. Tool names are non-standard (gdrive_connector lacks verb prefix; GPT5-* prefix is non-idiomatic). No output schemas documented. No error handling guidance. No parameter constraints (enums, ranges). Descriptions do not explain WHEN to use each tool or what happens on failure.
Legacy Codegen Support Tool - Uses GPT-5 to repair broken JavaScript code snippets for GDrive operations. When the AI-generated JavaScript code from ToolGDriveConnector fails to execute, this tool can be invoked to analyze the broken code and error message, then generate a corrected version.
Standalone Structured Tool - Uses GPT-5 to perform web searches and return results. Accepts a simple query parameter and delegates the web search to the JavaScript side, which performs the actual search and returns formatted results.
Legacy Codegen Tool - Manages GDrive connector tool calls between OpenAI WebRTC and JavaScript. The AI provides a self_contained_javascript_gdrive_code_snippet parameter containing executable JavaScript code, which is then forwarded to the JS side for execution.
All three tools lack parameter descriptions. Input schemas define types but omit descriptions for every parameter (self_contained_javascript_gdrive_code_snippet, task_description, broken_code, error_message, query).
Tool names do not follow verb_noun convention. 'gdrive_connector' lacks action verb (should be 'execute_gdrive_code' or 'run_gdrive_operation'). 'GPT5-gdrive-fixer' and 'GPT5-web-search' use non-standard prefixes and hyphens instead of underscores. LLMs rely on verb-first names to infer intent.
Descriptions are legacy-focused and lack actionable context. gdrive_connector describes 'Legacy Codegen Tool' and internal mechanics (OpenAI WebRTC, JavaScript forwarding) rather than WHAT the tool does for the user and WHEN to call it. No guidance on success/failure modes or next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No output schemas documented. Tool definitions show input schemas only. LLMs cannot plan downstream calls or extract results without knowing what fields are returned.
No error handling guidance. Descriptions do not explain what happens on failure, whether errors are retryable, or what the LLM should do next. Try search_users() with a partial name.'