The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external Zarr stores using AtomicDataZarrReader. It documentation discusses a skip_validation parameter that bypasses Pydantic validation for performance gains (~4 ms/sample). The skill includes a specific security warning advising users not to use this flag when store contents are untrusted or from a third party, effectively addressing the risk of processing malicious data structures.
[COMMAND_EXECUTION]: The skill provides benchmark examples using the nvalchemi-io-test CLI tool via uv run. These commands are standard for local performance testing of the data pipeline and are restricted to benchmarking storage and batching throughput without involving untrusted inputs or remote execution.

nvalchemi-zarr-perf