Files
hermes-agent/tests/tools
liuhao1024 ce963a46b1 fix(tts): play the artifact the tool reported in the sync speaker pipeline
text_to_speech_tool reports its artifacts (file_path/file_paths) in a JSON
envelope, and they often land beside — not at — the requested path: a command
provider's declared format rewrites the suffix, and voice-compatible delivery
ffmpeg-converts to .ogg. _synthesize_to_tmp ignored the envelope and returned
its own mkstemp path, so _drain's playback gate was false for every such
provider: voice mode was silent, nothing was logged, and the real audio files
leaked into TMPDIR. Prefer the first reported artifact that exists and is
non-empty, falling back to the requested path.

Fixes #115029
2026-09-20 10:25:30 -07:00
..