The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/package-loader.mjs script contains a dependency bootstrapping system that uses npm install to download required helper packages (@hyperframes/producer, @hyperframes/core) if they are missing from the local environment. While it includes mitigations like disabling lifecycle scripts and using a temporary directory, it still pulls code from a remote registry at runtime.
[DYNAMIC_EXECUTION]: The scripts/package-loader.mjs script utilizes dynamic import() to load and execute the bootstrapped npm packages. This pattern allows for the execution of code that was not part of the original skill package.
[INDIRECT_PROMPT_INJECTION]: The scripts/animation-map.mjs script implements an auditing capability that analyzes a user-specified directory (<composition-dir>). This directory is loaded into a headless browser session where its JavaScript is executed to performbbox sampling and metadata extraction. This represents an attack surface where malicious input content could influence the agent's execution environment.
Ingestion points: The audit script accepts a file system path to an untrusted directory containing HTML/JS compositions.
Boundary markers: None; the script expects a standard HyperFrames composition structure.
Capability inventory: The script executes arbitrary JavaScript via a browser session (createCaptureSession and page.evaluate) and performs file system writes to output the audit report.
Sanitization: No sanitization or sandboxing of the input composition code is performed prior to browser-based execution.

hyperframes-animation