The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The generated environment server processes CallToolAction objects, which contain tool arguments typically provided by a model. This creates a potential surface for indirect prompt injection if the inputs are not properly sanitized before execution in the environment.
Ingestion points: The step method in the generated server/<env>_environment.py file.
Boundary markers: The provided templates do not include explicit instructions for input validation or boundary markers.
Capability inventory: Depending on the archetype, capabilities include E2B sandboxes, browser automation, and database interactions.
Sanitization: No sanitization logic is present in the boilerplate code.
[EXTERNAL_DOWNLOADS]: The skill's Dockerfile instructions reference the ghcr.io/huggingface/openenv-base:latest base image from Hugging Face's official container registry.

generate-openenv-env