The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to download a setup script (setup.sh) from the vendor's repository on GitHub and execute it using the shell's source command. This pattern grants the external script full access to the agent's execution environment to configure the deployment.
[COMMAND_EXECUTION]: The skill extensively uses shell commands, including bash, docker compose, and curl, to manage container lifecycles and interact with the DataPrep REST API. This includes a piped command to python3 -c for parsing JSON health status, which is a safe handling of local service data.
[EXTERNAL_DOWNLOADS]: When a local clone is missing, the skill fetches deployment artifacts and documentation directly from the open-edge-platform/edge-ai-libraries repository on GitHub.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted data through the API.
Ingestion points: POST /summary (video_summary field) and POST /videos/upload (tags parameter) in SKILL.md and associated example prompts.
Boundary markers: None identified; the skill does not instruct the agent to ignore or delimit embedded instructions in the ingested metadata.
Capability inventory: The agent has access to bash, docker, and curl for system-level operations across multiple scripts.
Sanitization: No evidence of filtering, escaping, or validation of user-supplied text before it is stored or subsequently listed via GET /videos.
[CREDENTIALS_UNSAFE]: The deployment instructions require setting MINIO_ROOT_USER and MINIO_ROOT_PASSWORD. The skill appropriately uses placeholders and recommends shell exports rather than hardcoding secrets, following security best practices.

vdms-dataprep-user