The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external web pages, creating a potential surface for indirect prompt injection.
Ingestion points: The skill scrapes text and HTML content from ScienceDirect journal pages using the evaluate_script tool as described in SKILL.md.
Boundary markers: Extracted data is structured into a JavaScript object (result) and subsequently rendered via markdown templates, providing a clear separation between data and instructions.
Capability inventory: The skill has access to navigate_page and evaluate_script for browser interaction and data processing.
Sanitization: Basic string manipulation like .trim() and .substring() are used for formatting, though no robust sanitization of external text content is implemented.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic JavaScript execution to extract journal metadata and interact with the page.
An evaluate_script block in SKILL.md performs DOM traversal and data extraction at runtime.
The skill uses an initScript to redefine the navigator.webdriver property to evade bot detection during the navigate_page call.
Instructions include logic to detect and automate interactions with Cloudflare Turnstile if present, involving locating and clicking elements within an iframe context.

sd-journal-browse