The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and downloads pretrained model weights from the SwinTransformer GitHub repository (Microsoft Research). This is a well-known and reputable source for computer vision research artifacts used to initialize model training.
[COMMAND_EXECUTION]: The skill requires the Bash tool to execute TAO Toolkit commands (e.g., oneformer train, oneformer evaluate) and docker run sequences. These operations are essential for the skill's purpose and are documented to occur within scoped container environments.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection via the processing of untrusted external data.
Ingestion points: The agent context ingests untrusted data from dataset annotation files (JSON) and user-provided image directories, as well as configuration YAML files (referenced in SKILL.md and references/skill_info.yaml).
Boundary markers: Explicit boundary markers or instructions to ignore embedded commands in dataset files are absent.
Capability inventory: The skill possesses the capability to execute shell commands via the Bash tool (references/skill_info.yaml).
Sanitization: Input validation is expected to be performed by the oneformer CLI tool and TAO Toolkit libraries within the official NVIDIA container environment.

tao-train-oneformer