The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill specifies the use of the openreward and pydantic libraries, which are standard packages for this domain and are retrieved from the official PyPI registry during installation.
[INDIRECT_PROMPT_INJECTION]: The skill defines a protocol for environments to ingest task specifications, which creates an inherent attack surface where untrusted task data could potentially influence agent behavior.
Ingestion points: The task_spec dictionary used in server.py and the list of tasks defined in tasks.py.
Boundary markers: The templates do not specify mandatory delimiters, leaving boundary enforcement to the implementer's specific environment logic.
Capability inventory: Generated code includes @tool decorated functions in the Environment class which are callable by the agent.
Sanitization: The provided scaffolding does not include default sanitization for task data, which is typical for a low-level framework generator.
[SAFE]: The skill provides guidance on managing sensitive keys like E2B_API_KEY by recommending platform-specific secret storage instead of insecure local environment files.

generate-ors-env