The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/v2/run-platform.mjs dynamically assembles a JavaScript execution payload by concatenating multiple local files, including core.mjs, platform-specific adapters (e.g., bilibili.mjs), and dispatch.mjs. This combined source string is then executed as a Node.js script via the ego-browser CLI tool. While the source components are local to the skill, this runtime assembly and execution pattern is a form of dynamic execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted metadata from user-supplied package.json files, such as video titles and descriptions, which are interpolated into browser automation logic.
Ingestion points: Untrusted data enters the agent's context through the readPackage function in scripts/lib/content-package.mjs, which parses platform-specific metadata from a local JSON file.
Boundary markers: The skill performs field-level validation in scripts/lib/content-package.mjs (e.g., checking for character limits and forbidden hashtag formats), but it does not utilize explicit boundary markers or safety instructions to wrap the interpolated text when passed to the browser environment.
Capability inventory: The skill possesses significant capabilities, including persistent file system writes (scripts/v2/lib/job-store.mjs), browser process management (scripts/v2/run-platform.mjs), and local command execution (scripts/v2/publisher.mjs).
Sanitization: The skill mitigates basic injection risks by using JSON.stringify to escape user-supplied strings before they are embedded into the JavaScript fragments sent to the browser in scripts/v2/run-platform.mjs.

video-publisher