Tools and setup

Select capabilities from the recipient's actual environment. Commands below illustrate local file operations; replace filenames with the selected assets and inspect outputs. No installation or microphone access occurs by loading this Skill.

Supplied files and basic checks

Python 3.9+ runs the bundled read-only PCM WAV checker with no third-party dependencies:

python3 scripts/check_wav.py /path/to/take.wav

Run from this package directory. It accepts uncompressed integer PCM WAV with 8-, 16-, 24-, or 32-bit samples and prints JSON. It reports duration, channels, sample rate, per-channel peak/RMS dBFS, full-scale sample fraction, and the fraction of frames below a chosen amplitude threshold. It is not a pitch extractor, calibrated loudness meter, listening system, or singing score. The silence threshold defaults to -60 dBFS and can be changed with --silence-db -50. Recording context determines whether those values are useful. Unsupported compressed/float WAV or a damaged file produces an error and exit 2. It reads only the supplied file and never records, uploads, or changes it.

With separately installed FFmpeg, inspect a video:

ffprobe -v error -show_format -show_streams -of json karaoke.mp4

For a supported stream, remux audio without re-encoding:

ffmpeg -n -i karaoke.mp4 -map 0:a:0 -vn -c:a copy backing.mka

Matroska audio may not import into a particular DAW. Choose a compatible container for the actual codec, or decode into a derived PCM WAV; keep the original file and offsets. A PCM derivative for the checker can be made with:

ffmpeg -n -i input-audio.m4a -map 0:a:0 -c:a pcm_s24le take-check.wav

These commands do not validate lyrics, cleanliness, export rights, or perceptual quality. Use actual frame inspection and listening for those judgments. FFmpeg options are documented in its official manual.

Recorder and DAW operation

Use a recorder or DAW already available through the host's supported tools. Identify the live project, selected track, input/output devices, and current transport before changing them. Discover the current schema or app controls; tool names in documentation are not proof of availability. Operate under the host's focus, desktop, filesystem, and permission boundaries. Preserve other work, read back uncertain mutations before retrying, and verify stopped transport plus the new raw audio artifact at completion.

Logic Pro is an optional macOS route. It requires a recipient-owned installation, an audio device, the required microphone permissions, and a working control channel. MongLong0214/logic-pro-mcp is an independently maintained optional MCP implementation. Follow its current upstream installation and setup documentation for the installed version. Discover its exposed schemas and operation-specific requirements; verify handshake, target application, live project/track reads, and the intended transport route before recording. Supported app variants and permission requirements can change. Installing a binary does not prove any of these operations work on the recipient's host.

An app-bound Accessibility/API route can be used when available and permitted, with live controls and readback. Neither a stale project snapshot nor synthesized placeholder track data is a safe live target. Do not assume several server instances can share MIDI endpoints. Use the upstream ownership requirements. DAW controls, file diagnostics, audio listening, and pitch analysis are separate capabilities; report which actually worked. A logic_audio file-analysis result, if exposed, does not by itself establish reference-performance comparison.

Pitch and listening

Use an existing audio-capable model or player for listening only if it actually receives the audio. Optional analysis libraries include Parselmouth for Praat-based analysis and librosa for signal processing. They are not bundled or required; install through their documented setup only when the task needs that capability. Consult the installed version's pitch API and document algorithm, sample rate, hop size, pitch bounds, voicing confidence, and alignment. A successful import is not accuracy evidence on mixed music.

External tools retain their own licenses. No third-party implementation or media is copied into this package.