mirror of
https://github.com/usestrix/strix.git
synced 2026-09-29 01:41:30 +00:00
da472ca0d5
When interactive mode is enabled (TUI), the guard that prevents sending a trailing assistant message to the Anthropic API was incorrectly skipped. The Anthropic API rejects such requests with "This model does not support assistant message prefill." regardless of whether the caller is in interactive or non-interactive mode. Updated the condition to also apply the guard when `_is_anthropic()` is true, so Anthropic models are always protected while non-Anthropic models that may support assistant prefill continue to be unaffected in interactive mode. Added three unit tests covering all three branches of the new condition. Co-Authored-By: Octopus <liyuan851277048@icloud.com> |
||
|---|---|---|
| .. | ||
| agents | ||
| config | ||
| interface | ||
| llm | ||
| runtime | ||
| skills | ||
| telemetry | ||
| tools | ||
| __init__.py | ||
| conftest.py | ||