cognitive-calibration

原文涵盖的主题

• Cognitive Calibration: Music-Inspired Creative Exercise for Writing & Reasoning
• Overview
• Capability Requirements
• Dependencies
• Workflow

Cognitive Calibration provides an optional multimodal creative warm-up. Rather than relying on repetitive textual prompts, the agent listens to an anonymous, live-played, unquantized .mp3 composition from Michael Luchtan's Music For Robots (https://github.com/luchtan-GIT/m4r-data) via readfile, analyzes how the performance moves off the metronomic grid, and applies those acoustic principles to subsequent writing, essays, or architectural reasoning. Multimodal Audio Reading: Requires an agent harness with audio reading support (readfile on binary .mp3 files). If the harness 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 only Python 3 standard library modules. 1. Fetch an Anonymous Track into Secure Storage: (Withholds the track title and metadata to prevent keyword priming.) 2. Listen & Analyze: Call readfile on

这里只展示节选，请以完整原文和当前要求为准。