The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to interact with the host system and the hermes CLI for auditing purposes. Key operations include checking file permissions (ls -la ~/.hermes/), reviewing system configuration (hermes config show), and applying remediations such as changing file modes (chmod 600) or updating CLI settings (hermes config set). These actions are contextually appropriate for a security auditing tool.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection (Category 8) because it processes untrusted external data when auditing other skills via hermes skills audit and skillspector scan.
Ingestion points: External skill code and GitHub repositories provided for scanning.
Boundary markers: The instructions do not define specific text delimiters, but they establish a reporting structure to isolate findings.
Capability inventory: The skill has the ability to write to configuration files, change system permissions, and uninstall software.
Sanitization: No explicit sanitization or escaping of the audited code is defined in the instructions.
[SAFE]: The skill implements significant safety measures to mitigate its privileged access. It follows a 'Rule of Two' framework, ensuring that the combination of sensitive data access, untrusted content reading, and output channels is never fully automated without human oversight. Remediations are categorized by safety levels (Green/Yellow/Red), and the agent is strictly prohibited from making changes during unattended executions (Report Mode).

iris-security-audit