The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes content from external URLs provided in the arguments. This external content is untrusted and can contain instructions designed to manipulate the agent's logic or output.
Ingestion points: Untrusted data enters the agent context via the WebFetch tool using the $ARGUMENTS variable in SKILL.md.
Boundary markers: Absent. No delimiters or warnings are provided to handle embedded instructions in the fetched data.
Capability inventory: The skill has access to Bash, Task, and WebFetch tools as defined in SKILL.md.
Sanitization: Absent. No validation or sanitization processes are mentioned for external content.
[COMMAND_EXECUTION]: The skill is granted Bash and Task capabilities in the allowed-tools metadata. These tools allow for powerful command-line operations which could be misused if input is manipulated or if the agent is successfully injected.

geo