The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a potential surface for indirect prompt injection due to its core web-processing functionality.
Ingestion points: The skill ingests untrusted data from external websites via the analyze_page and compare_pages tools, which capture framework metadata, network request logs, and browser console output (SKILL.md).
Boundary markers: The instructions do not define boundary markers or provide the agent with explicit instructions to ignore embedded commands within the fetched external web content.
Capability inventory: Across its scripts, the skill provides capabilities to write persistent records via create_finding and perform follow-up network requests via get_observation and get_crawled_urls (SKILL.md).
Sanitization: There is no evidence of sanitization, filtering, or structured validation to ensure that text extracted from external sources like console logs is not interpreted as instructions by the agent.

web-research