The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from external Udemy course pages, which is a common surface for indirect prompt injection attacks.
Ingestion points: Scraped text from Udemy course URLs (title, description, curriculum) entered via the evaluate.js script.
Boundary markers: Absent. The extracted content is returned directly to the agent without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The skill itself only extracts text; it does not contain file-write or network-send capabilities, though it relies on the presence of a Chrome DevTools skill.
Sanitization: Minimal sanitization is performed via .trim() and .replace(), though a 3000-character limit is applied to the description field.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill references a local prerequisite skill and a standard npm install command within a local directory. It does not download or execute scripts from untrusted remote URLs.
[Dynamic Execution] (SAFE): JavaScript code is passed to the browser to extract data. This code is hardcoded for scraping logic and does not dynamically generate executable strings from untrusted external inputs.

udemy-crawler