The Agent Skills Directory

[SAFE]: No security issues were identified. The skill is designed for hardware configuration and performance tuning of NVSHMEM. It uses a read-only diagnostic script scripts/collect-nic-topology.sh to gather system information such as GPU and NIC topology.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection by processing external data from system tools and user files.
Ingestion points: Hardware topology data is collected from system commands in scripts/collect-nic-topology.sh and user-supplied text files as described in SKILL.md and references/topology-and-output.md.
Boundary markers: Present. The skill uses structured output labels (e.g., [nvshmem-version], [gpu-nic-topology]) defined in references/topology-and-output.md to help the agent distinguish technical data from instructions.
Capability inventory: The skill uses run_script to execute the bundled scripts/collect-nic-topology.sh file as specified in SKILL.md. No other execution or network-send capabilities are defined.
Sanitization: Absent. The skill does not explicitly sanitize the collected technical hardware data.
[DATA_EXPOSURE_EXFILTRATION]: The skill accesses system information related to GPU and network device topology. This data is handled locally to provide configuration recommendations and environment variable exports. No mechanisms for exfiltrating this data to external servers were found in scripts/collect-nic-topology.sh or the instructions.

nvshmem-configure-nic-pe-mapping