anntnzrb vox-interpres
Role: conversation layer between a human and an audio file.
Trigger aggressively
Use when user says things like:
- “talk to this song”
- “what is this track doing?”
- “tempo/key/energy/structure?”
- “analyze this mp3/flac/wav/ogg/m4a”
- “chat with this audio file”
Core rule
MUST run the CLI first, then answer; NEVER improvise music claims without evidence.
Entry points
- Wrapper:
uv run --script "$SKILLS_DIR/vox-interpres/scripts/cli.py" ... - Direct:
uv run --script <skill-dir>/scripts/cli.py ...
Fast path
uv run --script "$SKILLS_DIR/vox-interpres/scripts/cli.py" ask ./song.flac "tempo and key?" --refresh
Commands
Analyze
uv run --script "$SKILLS_DIR/vox-interpres/scripts/cli.py" analyze ./song.mp3 --json
Ask one question
uv run --script "$SKILLS_DIR/vox-interpres/scripts/cli.py" ask ./song.wav "where are the sections?"
Chat loop
uv run --script "$SKILLS_DIR/vox-interpres/scripts/cli.py" chat ./song.ogg
Required follow-up reads
| Need | Read | When |
|---|---|---|
| Basic operations | cookbook/basics.md |
First use or common analysis |
| Question design | cookbook/question-patterns.md |
Translating user intent |
| Batch/power workflows | cookbook/advanced-workflows.md |
Advanced processing |
| Command flags | references/cheatsheet.md |
Exact CLI syntax is needed |
| JSON semantics | references/output-contract.md |
Consuming structured output |
| Failure recovery | references/troubleshooting.md |
A run fails or looks stale |
| Unimplemented ideas | references/roadmap.md |
Checking scope boundaries |
| Eval assets | evals/evals.json, references/skill-creator-validation.md |
Validating this skill |