The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted YAML files and utilizes shell tools (Bash). This creates a potential attack surface where instructions embedded in a configuration file could influence agent behavior. However, the instructions emphasize structural validation and minimal edits, which serves as a mitigation.
Ingestion points: Reads and edits .yaml and .yml files (SKILL.md).
Boundary markers: Not explicitly defined for the prompt interpolation steps.
Capability inventory: Broad access to system tools via Bash(general:*) and Bash(util:*), along with file system read/write capabilities (SKILL.md).
Sanitization: The skill implements syntax and schema validation as its primary function, which helps identify and neutralize malformed data structures.
[SAFE]: No malicious patterns, obfuscated commands, or unauthorized data exfiltration attempts were detected. The skill provides helpful examples for managing secrets securely within infrastructure-as-code files.

yaml-master