audio-production

Machine interface: audio-production run - --json with a audio-production/request@1 document on stdin; exactly one audio-production/response@1 document on stdout. skill --json prints the contract, doctor --json the environment, plan - --json a dry run, validate - --json a schema check.

Rules for a calling agent:

1. Measure first with media-analysis-skill (or ffmpeg-skill probe / loudness --measure-only); decide; then build a request.
2. Give explicit, typed parameters: NORMALIZE needs target_lufs and true_peak_db; SILENCE_REMOVE needs the ranges you decided to remove. This skill has no defaults for production values and never guesses.
3. Never send commands, argv, filter strings or executable paths: the request is rejected.
4. Keep outputs inside the workspace, never at an input path; set overwrite: true deliberately.
5. Read results[].status, outputs[].provenance and error from the JSON; the exit code is 0 only when ok is true.

See README.md for the schemas, operation table, error codes and limitations.