You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(knowledge): give video-digest's ASR prerequisite a working install path and interpreter option #7743
A user can make the optional ASR rung work on a machine whose default Python refuses pip install --user (PEP 668, externally managed): a userConfig option names the interpreter that imports faster_whisper, detection tries it before the python/python3/py PATH candidates, SKILL.md Prerequisites carries a copyable uv venv recipe, and CUDA libraries installed as nvidia-* wheels load without PATH edits. (Gaps 2 and 3 of #6047.)
Recipe reported working by the producing run (not re-run): uv venv <dir> --python 3.13, then uv pip install faster-whisper nvidia-cublas-cu12 nvidia-cudnn-cu12 "av<19". The av<19 pin works around a reported faster-whisper 1.2.1 / PyAV 19 incompatibility (TypeError: open() got an unexpected keyword argument 'metadata_errors'); it was reported, not reproduced.
Constraints:
Never auto-install faster-whisper; ASR stays optional and detected at runtime.
Tests must not run a real Whisper model; stub runAsr and detection.
Bump the knowledge plugin version above whatever main holds at merge time and add a knowledge CHANGELOG entry. If vendored code under the knowledge plugin's vendor/ is touched, the vendor version-bump check applies too.
An interpreter path supplied through the new userConfig option is probed before the PATH candidates (unit test with a stubbed spawn).
With the option unset, detection behaves as today (test).
SKILL.md Prerequisites contains a copyable faster-whisper install recipe that includes the av<19 pin, with a recheck trigger for faster-whisper releases after 1.2.1.
asr_transcribe.py adds each existing site-packages/nvidia/*/bin directory as a DLL directory on Windows and is a no-op elsewhere.
Type: Bug Summary: Give video-digest's optional ASR rung a configurable interpreter, a working install recipe and Windows CUDA DLL loading.
Current behavior:
ASR detection tries only python, python3 and py from PATH. On a machine whose default Python refuses pip install --user (PEP 668), there is no supported way to point detection at a venv that imports faster_whisper. SKILL.md gives no working install recipe, and CUDA libraries installed as nvidia-* wheels do not load on Windows without PATH edits.
Desired behavior:
A new knowledge userConfig option names an interpreter; when set, detection probes it before the PATH candidates. Unset, detection behaves as today.
SKILL.md Prerequisites renders the option in the SKILL.md body (not a spoke) and carries a copyable uv venv recipe: uv venv <dir> --python 3.13, then uv pip install faster-whisper nvidia-cublas-cu12 nvidia-cudnn-cu12 \"av<19\", with a recheck trigger for faster-whisper releases after 1.2.1. Nothing is auto-installed.
The ASR runner script adds each existing site-packages/nvidia/*/bin directory as a DLL directory on Windows only, and does nothing elsewhere.
Key interfaces:
detectAsrCapability({ spawn }) and its interpreter candidate order.
Knowledge plugin manifest userConfig: one new interpreter-path option.
ASR runner script: Windows DLL directory setup.
Acceptance criteria:
A vitest case with a stubbed spawn shows a configured interpreter is probed before the PATH candidates, and another shows unchanged behavior with the option unset; scripts/run-test-file.sh on that file exits 0.
SKILL.md Prerequisites contains the recipe with the av<19 pin and a recheck trigger, and renders the option in the body (read from the diff).
The runner script adds the DLL directories only when running on Windows and only for directories that exist (read from the diff).
A knowledge changelog fragment is added and scripts/check-changelog-fragments.sh exits 0.
Out of scope:
Auto-installing anything, or running a real Whisper model in tests.
This was generated by AI during backlog shaping.Held, not admitted: this brief is drafted and checked, but the issue lane would stop on it today. It is admitted, by adding agent-ready, once the lane can run the skill-quality check scripts (Automation Lanes' PR #7912). Held under the owner's 2026-10-18 lane-narrowing course correction.
Activity
kyle-sexton commentedon Oct 17, 2026
Ratified in the owner's batch of decomposed children (2026-10-17).
kyle-sexton commentedon Oct 18, 2026
Agent Brief
Type: Bug
Summary: Give video-digest's optional ASR rung a configurable interpreter, a working install recipe and Windows CUDA DLL loading.
Current behavior:
ASR detection tries only
python,python3andpyfrom PATH. On a machine whose default Python refusespip install --user(PEP 668), there is no supported way to point detection at a venv that importsfaster_whisper. SKILL.md gives no working install recipe, and CUDA libraries installed asnvidia-*wheels do not load on Windows without PATH edits.Desired behavior:
uvvenv recipe:uv venv <dir> --python 3.13, thenuv pip install faster-whisper nvidia-cublas-cu12 nvidia-cudnn-cu12 \"av<19\", with a recheck trigger for faster-whisper releases after 1.2.1. Nothing is auto-installed.site-packages/nvidia/*/bindirectory as a DLL directory on Windows only, and does nothing elsewhere.Key interfaces:
detectAsrCapability({ spawn })and its interpreter candidate order.userConfig: one new interpreter-path option.Acceptance criteria:
scripts/run-test-file.shon that file exits 0.av<19pin and a recheck trigger, and renders the option in the body (read from the diff).scripts/check-changelog-fragments.shexits 0.Out of scope:
This was generated by AI during backlog shaping. Held, not admitted: this brief is drafted and checked, but the issue lane would stop on it today. It is admitted, by adding
agent-ready, once the lane can run the skill-quality check scripts (Automation Lanes' PR #7912). Held under the owner's 2026-10-18 lane-narrowing course correction.test(knowledge): cover the Windows nvidia DLL-directory step in asr_t…