The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Several workflow configuration files (e.g., assets/configs/finetune.yaml, assets/configs/good_image_generation.yaml) download the yq utility directly from its official GitHub repository (github.com/mikefarah/yq) during task execution to manipulate configuration files in-pod.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including CAD USD files, real optical photos, and validation metrics from user-defined storage roots. It also instructs the agent to monitor workflow logs for status updates.
Ingestion points: External datasets, validation CSVs (valid_kpi.csv), and real-time task logs are ingested into the agent's context.
Boundary markers: Absent; external data and log outputs are not enclosed in delimiters.
Capability inventory: The skill utilizes Shell and Read tools to execute osmo CLI commands, manage filesystem assets, and monitor logs.
Sanitization: No explicit sanitization or filtering of external data content is implemented prior to processing.
[COMMAND_EXECUTION]: The skill uses various shell scripts (scripts/preflight_credentials.sh, scripts/preflight_urls.sh, etc.) and inline shell commands within YAML configurations to orchestrate the Defect Image Generation (DIG) pipeline, including environment setup and workflow submission.

physical-ai-defect-image-generation