| # Downloaded files | |
| Agent's requirements for the downloaded files: | |
| - **Detect the file type** based on the extension and the linked question (referenced in `common_questions.py`). | |
| - **Audio Files (.mp3):** | |
| Use speech-to-text routines to transcribe the audio and then extract the required answer (e.g. page numbers or ingredient lists) with proper post‑processing (sorting, deduplication). | |
| - **Spreadsheet Files (.xlsx):** | |
| Parse the Excel file (using tools like pandas) to extract the numeric sales data, summing up relevant columns and formatting the number as specified. | |
| - **Image Files (.png):** | |
| Apply OCR or vision reasoning to interpret the chess position (or extract text details) so that the agent can compute the correct algebraic move. | |
| - **Python Code Files (.py):** | |
| Safely execute the code snippet in a sandboxed environment to capture its output (the final numeric value). | |
| In short, the agent must determine which specialized tool (speech-to-text, spreadsheet reader, OCR/vision, or code interpreter) to invoke for each file type and then synthesize the outputs, ensuring that the final answers meet the exact-match formatting specified in the submission instructions (submission_instructions.md). |