The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves an agent processing untrusted source code files, creating a vulnerability surface for indirect prompt injection attacks. 1) Ingestion points: The target file list is determined by user-supplied flags as described in the Scope Determination workflow. 2) Boundary markers: The instructions lack explicit requirements for using delimiters or 'ignore' warnings when reading code content. 3) Capability inventory: The skill directs an agent to read source code files and write analysis findings to the file system. 4) Sanitization: There is no mention of sanitizing or escaping content from the analyzed files before it is processed by the agent.

dos