| name | minnarone-runtime-doctor |
| description | Diagnose a Minnarone runtime profile before shadow. Use to check extras, tools, pinned models, config, lazy model probes and Twitch safety gates without downloading weights or starting live. |
Minnarone runtime doctor prototype
Read the profile matrix and pinned manifest before checking a machine. Be
read-only by default: do not install tools, download/delete weights, start a
stream or contact Twitch without a separate explicit request.
Checks
- Select P0, P2, or one full profile P3-P5; do not assume “all extras”.
- Report PASS/FAIL/SKIP for Python,
uv, profile extras, Streamlink, FFmpeg,
free disk, model paths, byte sizes, revision and SHA-256.
- For Italian audio require CAM++ English VoxCeleb
dimension: 512; reject a
zh-cn 192 model/config mismatch. Treat threshold 0.5 as a starting point.
- Make lazy boundaries explicit:
--check does not prove a Qwen load, usable
CUDA/MPS memory, or the first caption.
- Verify dedicated bot account, broadcaster-consent status, disclosure choice,
retained artifacts/manual deletion and
send.mode: shadow.
- Propose isolated smoke commands, then
minnarone <config> --check and a
bounded shadow run. Never press p or promote live.
The disposable ../../doctor.py and ../../manifest.json exercise the
contract; canonical profiles live in
../../../../docs/research/runtime-model-profiles.md. Production should keep
docs canonical and let the skill orchestrate checks; a core minnarone doctor
command remains deferred.
