mirror of
https://github.com/Significant-Gravitas/AutoGPT.git
synced 2026-02-05 12:25:04 -05:00
When blocks produce binary outputs (PNG, JPEG, PDF, SVG), the data is now automatically saved to the user's workspace and replaced with workspace:// references. This prevents: - Massive token waste from LLM re-typing base64 strings (17,000+ tokens) - Potential data corruption from truncation/hallucination - Poor UX from slow character-by-character output Implementation: - New binary_output_processor.py module with hash-based deduplication - Integration in run_block.py (single entry point for all block executions) - Graceful degradation: failures preserve original data Fixes SECRT-1887