Single tool 'offload' has a clear verb-noun name and reasonable description (156 chars), but the input schema is problematic. Parameters lack individual descriptions, only the top-level schema shows brief hints. The 'task' parameter accepts a constrained set of values but is documented as a free-form string rather than an enum, forcing the LLM to memorize valid options. The 'source' parameter similarly lacks enum constraint. Output schema is completely undocumented, callers have no visibility into what fields the response contains. Error handling is minimal; the tool provides no recovery guidance. The implementation shows good internal validation and rate-limiting logic, but none of that surfaces in the tool definition.

• Convert 'task' parameter to an enum with values: commit_message, pr_description, code_summary, translate, changelog_entry, naming_suggestion, classify, extract_data, code_review_single, docstring, subject_lines, freeform. Add description: 'Type of task to offload. Use freeform for custom prompts.'
• Convert 'source' parameter to an enum with values: git_diff, git_staged_diff, file. Add description: 'Source of content: git_diff (HEAD changes), git_staged_diff (staged changes), or file (read from path).'
• Add individual parameter descriptions: path='File path (required if source=file)', cwd='Working directory for git commands (defaults to current directory)', prompt='Custom prompt text (required if task=freeform)'.
• Document output schema in tool description: 'Returns {text: string, totalTokens: number, model: string}. text contains the LLM response; totalTokens is the token count used; model is the model that generated the response.'
• Enhance error messages with recovery guidance. E.g., 'API key not configured: set GOOGLE_AI_API_KEY environment variable' or 'Unknown task: must be one of commit_message, pr_description, ... or freeform.'