Scoring was not performed

• Document the output schema for every tool. For upload_pdf, specify what fields are returned (e.g., document_id, indexed_chunks_count, status). For chat, document the structure of the response (e.g., answer, sources, confidence_score). For workflow tools, document what fields a workflow object contains (e.g., workflow_id, status, created_at, results).
• Expand tool descriptions to 100-200 characters following the pattern: 'What does it do?' + 'When to use it?' + 'Key dependencies'. Example: 'Run a workflow synchronously and wait for the complete result. Use this for small, quick analyses. For long-running analyses, use stream_workflow instead. Requires a valid workflow_id from list_workflow_runs or previous runs.'
• Add contextual descriptions to workflow_id, ticker, and company_name parameters. Explain: Is workflow_id a UUID or integer? Where does it come from? Must it be pre-created? Example: 'Unique identifier for the workflow run (UUID format). Obtain from list_workflow_runs() or create a new analysis by specifying workflow_id as a unique string.'
• For list_workflow_runs, document the return structure. Specify: 'Returns a paginated list of workflow run objects, each containing: workflow_id, ticker, company_name, status, created_at, updated_at, results_summary. Use cursor for pagination.'
• Add error recovery guidance to chat and run_workflow: 'If the query times out or returns no results, try: (1) reducing top_k parameter; (2) broadening company_name or domain filters; (3) uploading additional documents with upload_pdf.'