The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is primarily designed to execute shell commands and scripts within isolated Modal Sandbox containers using the official modal SDK, as demonstrated in scripts/smoke_test.py.- [REMOTE_CODE_EXECUTION]: It facilitates the execution of user-supplied code on remote Modal infrastructure. All remote interactions are conducted through the authenticated Modal SDK, which is a well-known cloud service.- [EXTERNAL_DOWNLOADS]: The skill requires and references the modal Python package. These dependencies are standard for interacting with the Modal platform and do not involve untrusted third-party script execution.- [PROMPT_INJECTION]: While the skill processes user-supplied code, it relies on the isolation of the sandbox environment to mitigate risks. There are no patterns suggesting attempts to bypass the agent's core safety guidelines.
Ingestion points: User-provided commands and scripts are passed to modal.Sandbox.create and sandbox.exec in SKILL.md and evals/evals.json.
Boundary markers: Instructions emphasize the use of isolated sandboxes for "untrusted or user-supplied code" to maintain environment security.
Capability inventory: The skill provides full execution capabilities within the container, including filesystem access (sandbox.open) and network tunnels (sandbox.tunnels) as documented in the reference files.
Sanitization: The skill does not perform string sanitization, instead relying on the infrastructure-level isolation of the Modal Sandbox container.

modal-sandbox