The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external codebase files, which are considered untrusted data. A malicious codebase could contain comments or strings designed to deceive the auditing agent.
Ingestion points: The skill reads file content from the directory path provided in $ARGUMENTS using tools like Read, Glob, and Grep.
Boundary markers: There are no explicit instructions or delimiters defined to separate the untrusted code content from the agent's internal analysis instructions.
Capability inventory: The skill is granted access to high-privilege tools including Bash and Task. If an indirect injection is successful, these tools could be leveraged to perform unauthorized actions on the host system.
Sanitization: The skill lacks instructions for sanitizing or escaping content read from the target codebase before it is processed by the agent.

auth-assessment