The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes untrusted data from external websites (robots.txt, HTML meta tags, and HTTP headers) using the WebFetch tool. An attacker who controls a site being analyzed could embed malicious instructions in these fields to attempt to influence the agent's behavior.
Ingestion points: External URLs provided by users to be processed by the instructions in SKILL.md (Analysis Procedure).
Boundary markers: The instructions lack explicit warnings to treat fetched content strictly as data or delimiters to separate data from instructions during parsing.
Capability inventory: The skill has access to Bash, Write, Read, Grep, Glob, and WebFetch tools as defined in the SKILL.md frontmatter, which could be abused if an injection is successful.
Sanitization: No sanitization or validation of the fetched content is described in SKILL.md before parsing and inclusion in the final report.
[COMMAND_EXECUTION]: The skill includes Bash in its allowed-tools list. While the instructions primarily focus on parsing data, the availability of a shell environment alongside tools that fetch external data increases the potential impact of an instruction injection from an untrusted web source.

geo-crawlers