The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to execute dmesg using sudo for driver debugging and notes that access to hardware devices (doca_dev) typically requires elevated privileges.
[COMMAND_EXECUTION]: Instructions guide the agent to run various system commands for environment discovery and building, including pkg-config, doca_caps, ls, and dmesg.
[INDIRECT_PROMPT_INJECTION]: The skill has the agent ingest data from system command outputs and local files (e.g., sample code in /opt/mellanox/doca/), creating a surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: Command outputs from dmesg, pkg-config, and doca_caps; local files in /opt/mellanox/doca/ (TASKS.md).
Boundary markers: None explicitly defined for untrusted command output.
Capability inventory: Subprocess calls for pkg-config, doca_caps, dmesg, ls, and which (TASKS.md).
Sanitization: None described for external command output.
[OBFUSCATION]: The skill includes an artifact signature file (skill.oms.sig) containing a Base64-encoded payload that embeds internal URLs not visible in plain-text scanning.

doca-pcc