The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests DNA sequences from user prompts for genomic model processing. Ingestion point: sequence variable in SKILL.md. Boundary markers: Output is formatted into structured JSON and FASTA files. Capability inventory: Utilizes requests.post for API interaction and docker run for local deployment. Sanitization: Employs a clean_dna function to enforce a strict A/C/G/T character whitelist, mitigating risks of indirect injection via malformed sequences.
[DATA_EXFILTRATION]: Sequence data and inference parameters are transmitted to NVIDIA's health API endpoints for processing. All network activity targets verified vendor infrastructure or local developer-managed instances.
[COMMAND_EXECUTION]: The skill provides documentation for Docker-based deployment, including registry login and container management commands.
[EXTERNAL_DOWNLOADS]: Local deployment involves pulling containerized model assets and weights from the official NVIDIA Container Registry (nvcr.io).
[DYNAMIC_EXECUTION]: Genomic tensors are processed using the numpy library. The skill explicitly disables insecure deserialization by setting allow_pickle=False when loading NPZ files, preventing potential code execution from malicious binary data.

evo2-nim