The Agent Skills Directory

[SAFE]: No security issues detected across any of the 11 threat categories. The skill's structure and instructions are purely informational and follow best practices for resource lookup.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads untrusted data from assets/data.csv via the read_skill_resource tool.
Boundary markers: Absent. No specific instructions are provided to the agent to disregard potential instructions embedded within the CSV data.
Capability inventory: None. The skill does not include scripts or instructions that trigger shell commands, network access, or file system modifications.
Sanitization: Absent. The CSV is read as raw text.
Analysis: Although the skill lacks explicit boundary markers for untrusted data, the total absence of exploitable capabilities (RCE, data exfiltration, or persistence) means this ingestion surface poses no practical security risk.

combined-resources