The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a framework-specific CLI tool (omh) to record runtime evidence. This behavior is consistent with the skill's stated purpose as a workflow guide within the Oh-My-Hermes (OMH) ecosystem.
Evidence: omh runtime record --skill gateway-intent-card --harness gateway-intent-card --status started in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external user requests and format them into structured gateway policies, creating a potential attack surface for indirect prompt injection if the user input contains malicious instructions.
Ingestion points: Untrusted data enters via the user request field as defined in the Required inputs section of SKILL.md.
Boundary markers: The skill explicitly mentions maintaining a prepared-vs-observed boundary to separate guidance from external data.
Capability inventory: The skill has the capability to execute shell commands via the omh tool and potentially interact with file system paths via the MEDIA:/absolute/path protocol.
Sanitization: The skill provides logical sanitization instructions by warning the agent not to claim external evidence (like message delivery) from prepared guidance, effectively limiting the agent's trust in its own generated output vs observed reality.
[DATA_EXPOSURE]: The skill mentions a specific protocol for handling file attachments using absolute paths (MEDIA:/absolute/path). While this identifies a pattern of file system awareness, it is used here as a format specification for reporting attachments rather than a command to access sensitive files.

omh-gateway-intent-card