The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data (support tickets) to generate content, which presents a surface for indirect prompt injection.
Ingestion points: Support ticket threads, resolutions, and customer conversations provided as input or retrieved via placeholders (e.g., ~~support platform in SKILL.md).
Boundary markers: The instructions lack explicit delimiting (such as XML tags or clear separators) to encapsulate and isolate external ticket content from the system instructions.
Capability inventory: The skill uses Read, Glob, and Grep tools to retrieve file-based information.
Sanitization: No specific instructions for filtering or sanitizing instructions potentially embedded within the ingested ticket data are provided.
[SAFE]: No obfuscation techniques such as Base64 encoding, zero-width characters, or homoglyphs were detected within the documentation or logic.
[SAFE]: The skill does not perform unauthorized network operations, exfiltrate sensitive data, or attempt to acquire elevated privileges.

kb-article