The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill performs runtime installation of the @strudel/web package from the NPM registry into a user-specific cache directory (~/.cache/listening-to-music/) using shell commands. It also dynamically evaluates user-supplied JavaScript and HTML content within a headless Playwright browser environment to facilitate music rendering and recording.
[COMMAND_EXECUTION]: Several scripts execute shell commands to manage environment settings and invoke external processing utilities. common.mjs runs npm root -g and npm install, while read_score.py utilizes subprocess.run to call oemer for optical music recognition and pdftoppm for image extraction from PDFs.
[EXTERNAL_DOWNLOADS]: The skill fetches audio samples and library code from external sources, including the NPM registry and public GitHub repositories (such as felixroos/dough-samples and todepond/samples), to provide the necessary assets for music synthesis.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data, such as music notation files and web content, creating a surface for indirect prompt injection due to its execution capabilities.
Ingestion points: Untrusted data enters via read_score.py (MIDI, MusicXML, and images), record_page.mjs (URLs/HTML), and render_strudel.mjs (JavaScript music code).
Boundary markers: No explicit delimiters or safety warnings are implemented to isolate instructions that might be embedded in processed music files or web data.
Capability inventory: The skill uses subprocess.run, execSync, and headless browser execution, which could be leveraged if malicious instructions were successfully injected.
Sanitization: Input data and file paths are not rigorously sanitized or validated before processing by external tools or evaluation in the browser environment.

listening-to-music