mechanical-ear-critic

原文涵盖的主题

• The Mechanical Ear Critic: Music-Inspired Structural Critique
• Overview
• Capability Requirements
• Dependencies
• Workflow

Inspired by The Mechanical Ear (by composer Michael Luchtan), where two distinct machine perspectives—Vincent Van Goghbot (attentive to what gets lost in compression and abstraction) and JS Robach (attentive to formal structures and edge cases)—review compositions in real time. Before reviewing the user's deliverable, the agent fetches an anonymous .mp3 composition from Music For Robots (https://github.com/luchtan-GIT/m4r-data) into a secure private directory, listens via readfile, analyzes how the performance handles tension, decay, and physical friction, and applies that calibrated ear to critiquing code, writing, or system designs. Multimodal Audio Reading: Requires an agent harness with audio reading support (readfile on binary .mp3 files). If the environment lacks audio reading tools, do not invoke this skill. No Metadata-Only Claims: Agents must only claim to have listened to the music if readfile was actually executed on the .mp3. Inspecting JSON manifests is text analysis, not listening. Private Unique Storage: Downloads use private, process-unique temporary directories (mode 0700) and user-private cache (/.cache/m4r/). uv: Used to run scripts/m4ringest.py (uv run). Uses on

这里只展示节选，请以完整原文和当前要求为准。