The Agent Skills Directory

[DATA_EXPOSURE]: The skill manages user-defined interests and feedback using a local storage directory at ~/.missav/. It reads and writes to preferences.json and feedback.jsonl to persist scoring logic across sessions.
[COMMAND_EXECUTION]: Local shell and Node.js scripts coordinate the interaction with the agent-browser tool. These scripts handle session management, data normalization, and ranking calculations within the skill's environment.
[EXTERNAL_DOWNLOADS]: The skill browses missav.ws to retrieve video listings. It implements defensive patterns by checking for Cloudflare challenges and ad-redirect URLs (e.g., /pop, tsyndicate), terminating or skipping as required to avoid bypassing access controls or visiting malicious domains.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata, such as titles and plot descriptions, from the target website. This content is used as input for local ranking scripts and summarized for the user, representing a common threat surface for browsing-capable skills, though no active exploitation is present in the instruction set.

missav