The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection through its core functionality of fetching and analyzing untrusted web content.
Ingestion points: The skill uses the WebFetch tool in Steps 1, 2, and 3 to retrieve data from external domains, including robots.txt files, HTML source code, and HTTP response headers.
Boundary markers: The instructions lack requirements for boundary markers or specific delimiters to isolate the fetched external content from the agent's logic, increasing the risk that the agent may interpret embedded text as instructions.
Capability inventory: The skill is granted powerful capabilities including Bash (shell execution), WebFetch (network access), and Write (file system modification) in SKILL.md, which could be misused if the agent is compromised by an injection attack.
Sanitization: There are no instructions for sanitizing, validating, or escaping the external data before it is processed by the agent or passed to tools like Grep or Bash.

geo-crawlers