The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is highly susceptible to indirect prompt injection because it ingests untrusted data and possesses write capabilities.
Ingestion points: The $1 argument (file path or topic description) and the contents of the file read via the Read tool in the 'Detecting Input Type' section.
Boundary markers: Absent. The instructions do not define delimiters or provide 'ignore embedded instructions' directives when processing external content.
Capability inventory: Includes reading local files, interacting with users via the AskUserQuestion tool, and most critically, updating files in-place ('update the file in-place with refined insights').
Sanitization: None. There is no logic to validate or escape the content being read before it is analyzed and used to determine the next agent actions.
[File System Manipulation] (MEDIUM): The instruction to 'update the file in-place' grants the agent the ability to overwrite data. An attacker could use a malicious file to trick the agent into corrupting sensitive files or injecting malicious content into existing scripts.

interview