The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content when auditing and redesigning websites, creating a surface for indirect prompt injection attacks.
Ingestion points: The snapshot.mjs script fetches HTML, CSS, and images from user-provided remote URLs.
Boundary markers: There are no explicit delimiters or warnings to prevent the agent from following instructions potentially embedded within the ingested web content.
Capability inventory: The skill possesses significant capabilities including file system writes, shell command execution (via execFileSync), and full browser automation (via Playwright).
Sanitization: The snapshot.mjs script includes a sanitization step that removes <script> and <noscript> tags from the captured DOM before processing.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Node.js scripts (snapshot.mjs and story.mjs). These scripts interact with the host system to launch browser instances and run media encoding processes via ffmpeg.
[DYNAMIC_EXECUTION]: The story.mjs script dynamically generates JavaScript 'runtime' strings and HTML templates which are then injected into a browser context to facilitate the creation of the video morphing effect.
[DATA_EXFILTRATION]: The skill performs network requests to arbitrary external domains to snapshot websites. Furthermore, the snapshot.mjs script does not validate the URL protocol, which could potentially be exploited to access local system files using the file:// scheme if the agent is manipulated.

unslop