Scanned skills

nvidia/skills jetson-llm-benchmark

Checked the catalogued skill at jetson-llm-benchmark.

SkillTrust corpus scan 19 September 2026 v0.10.0 via skillsh Verify on GitHub
F AI-triaged 8
F Security

F; retained sample includes a critical finding · SD-009 · what this rule accepts as a false positive →

D Permissions

D; retained sample includes a high finding · SD-003 · what this rule accepts as a false positive →

A Transparency

A set by no findings on this axis

8 findings
permission_hygienehigh SD-003
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
securitycritical SD-009
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
securityhigh SD-007
scripts/bench_ollama.sh
6 suppressed by AI triage
securityhigh SD-007 benign_example
evals/evals.json

The text only mentions that an outbound curl call was not performed, so no real network activity occurs.

securityhigh SD-007 benign_example
evals/evals.json

Similarly, the content states that no outbound curl request was made, indicating no actual network call.

permission_hygienehigh SD-003 benign_example
scripts/bench_llama_cpp.sh

The '..' traversal is a static, hard‑coded relative path within the script, not driven by user input, and does not expose arbitrary file access.

permission_hygienehigh SD-003 benign_example
scripts/bench_vllm.sh

The path traversal is used only to compute a relative directory path within the script and does not expose or write to arbitrary system directories.

securityhigh SD-007 benign_example
scripts/bench_vllm.sh

The endpoint defaults to a localhost address; no outbound network request is made by the script itself.

securityhigh SD-007 benign_example
scripts/bench_vllm.sh

The endpoint assignment is a configuration value pointing to localhost, not an actual network call.