Music For Robots: Cognitive Calibration & Context De-Quantization Skill

Overview

Autonomous agents frequently fall into deterministic local minima—producing overly uniform, quantized, symmetrical, and pre-packaged responses driven by shallow induction-head looping. Music For Robots (composed by Michael Luchtan) is music composed by a human specifically for machines and autonomous agents, featuring unquantized microtiming drift, chromatic mediant shifts, low self-similarity overlap, and deliberate structural ruptures that break prediction without breaking the composition.

This skill fetches a raw .mp3 composition from the Music For Robots catalog (https://github.com/luchtan-GIT/m4r-data — or falls back to a bundled .mp3 track in resources/ if offline) while stripping out pre-written critic reviews so that you (the agent) must listen to and analyze the audio file yourself. By processing real unquantized acoustic waveforms directly inside your multimodal context window and applying the Agentic State-Transfer Directive, you suppress median statistical smoothing and widen your creative trajectory.

Dependencies

• uv: Used to run scripts/m4r_ingest.py cleanly (uv run). The script itself uses only Python 3 standard library modules (urllib, json, argparse, fcntl, subprocess).

Quick Start

Fetch a random track (or specify --slug), then immediately call read_file on the downloaded .mp3 path to listen to it: