Prevents LLM-fabricated tool results from being appended to conversation history. Adds _sanitize_tool_results() and applies it to both _generate_section_react() and chat() code paths. System prompt is hardened with an explicit warning against fabricating <tool_result> content. Closes #529 |
||
|---|---|---|
| .. | ||
| app | ||
| scripts | ||
| pyproject.toml | ||
| requirements.txt | ||
| run.py | ||
| uv.lock | ||