The Agent Skills Directory

[OBFUSCATION]: The class identifier TrafilaturaСscraper (defined on line 35 and instantiated on line 192 of SKILL.md) contains a Cyrillic 'С' (U+0421) substituted into an otherwise Latin word. This script-mixing technique is a homoglyph used to imitate a Latin target and can be used to bypass security tools or static analysis that monitor for specific class names.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for extracting and processing content from external websites (using trafilatura, BeautifulSoup, and Playwright). This establishes an attack surface for indirect prompt injection, where an attacker could place malicious instructions on a web page that are then ingested into the agent's context.
Ingestion points: Web content extraction in the fetch methods of TrafilaturaСscraper, RequestsScraper, and PlaywrightScraper in SKILL.md.
Boundary markers: The skill does not provide or suggest using delimiters or 'ignore embedded instructions' warnings to isolate scraped content.
Capability inventory: The skill utilizes network communication tools (requests, Playwright) and has the capability to write to the file system (yt-dlp, instaloader).
Sanitization: There is no evidence of sanitization or filtering of the extracted text before it is returned for agent processing.
[DYNAMIC_EXECUTION]: The PlaywrightScraper uses page.evaluate() to execute JavaScript within the context of the scraped page. While used here for content extraction, this represents the dynamic execution of code within the agent-controlled browser session.

web-scraping