The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/lib/browser.mjs implements dynamic module loading by using import() on paths resolved at runtime via createRequire().resolve(). It specifically attempts to load playwright or playwright-core from the current working directory or the skill's own directory structure, which allows for execution of code from computed paths.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection as it ingests untrusted data from external websites.
Ingestion points: The loadUrlSite function in scripts/lib/pages.mjs fetches HTML content from user-provided or crawled URLs.
Boundary markers: The skill does not implement specific delimiters or instructions to the agent to disregard instructions embedded within the extracted metadata (such as titles, descriptions, or Open Graph tags) returned by the scripts.
Capability inventory: The skill executes shell commands to run its analysis scripts and performs both network requests to external domains and local file system reads (when running in directory mode).
Sanitization: While it decodes HTML entities in scripts/lib/html.mjs, it does not perform sanitization to strip potential adversarial instructions before the data is presented to the agent.
[COMMAND_EXECUTION]: The SKILL.md file instructs the agent to run local Node.js scripts, such as inspect-metadata.mjs, using its shell tools. These scripts are granted capabilities to interact with the file system and perform network operations, which could be leveraged if malicious arguments were successfully injected into the command invocation.

multilingual-readiness