The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data fetched from external websites, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: The fetch_domain function in scripts/check.py retrieves content from the target domain's homepage, robots.txt, llms.txt, pricing, and careers pages.
Boundary markers: While the script returns structured JSON or formatted text, there are no explicit delimiters or specific instructions for the agent to ignore natural language instructions embedded within the fetched external content.
Capability inventory: The skill possesses the Read tool, Write (restricted to the ./gtm/ directory), and Bash capabilities (specifically for executing the bundled check.py script).
Sanitization: The PageParser class in scripts/check.py extracts text and meta information while ignoring script and style tags, but it does not perform sanitization to filter out malicious natural language instructions that might attempt to influence the agent's behavior.

gtm-leak-check