Follow-up polish from review, both only reachable when the new flag is set: - gguf loop: suppress the streamed provisional "running" card for a python/terminal call when confirm_code_execution gates it, so a large code-execution call no longer flashes a card before the approve/deny prompt (mirrors the confirm_tool_calls suppression). - routes: reject confirm_code_execution for Anthropic Messages server tools with a 400, matching the existing confirm_tool_calls rejection, instead of silently ignoring it. |
||
|---|---|---|
| .. | ||
| data_recipe | ||
| export | ||
| inference | ||
| rag | ||
| training | ||
| __init__.py | ||
| _torchao_stub.py | ||
| import_guards.py | ||
| tool_healing.py | ||