The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the user to execute diagnostic commands using root privileges through the use of sudo to inspect system logs and hardware state.
Evidence: TASKS.md specifies the use of sudo dmesg | tail -n 40 for checking kernel and driver logs.
Evidence: TASKS.md specifies the use of sudo ibv_devinfo to verify the state and configuration of the InfiniBand/RDMA hardware substrate.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow ('modify') that involves the agent ingesting and acting upon untrusted user descriptions of high-performance computing (HPC) patterns and memory operations.
Ingestion points: The modify workflow in TASKS.md and SKILL.md requires the agent to elicit details from the user regarding operation families, memory descriptors, and HPC stack integration.
Boundary markers: The skill lacks instructions for the agent to utilize delimiters or safety warnings to mitigate the risk of processing malicious instructions embedded within the user's application logic.
Capability inventory: The skill possesses the capability to guide the user in generating, compiling, and executing C/C++ code and shell scripts, creating a potential path for exploitation if the agent follows malicious instructions in the user's input.
Sanitization: There is no evidence of sanitization or strict schema validation for the application-specific details provided by the user.
[COMMAND_EXECUTION]: The skill provides various shell commands intended for environmental configuration and verification.
Evidence: TASKS.md includes commands for pkg-config --modversion doca-urom, doca_caps --list-devs, and file system operations to locate version files and sample code.

doca-urom