The Agent Skills Directory

[SAFE]: The skill is a mock implementation intended for development and testing. It does not invoke any interpreters, compilers, or subprocesses, and explicitly returns synthetic output as defined in the _simulate_execution function in tools.py.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code snippets via tool parameters, representing a theoretical attack surface.
Ingestion points: The code parameter in agent.py and the execute_code/validate_code tools in tools.py.
Boundary markers: None implemented; user input is passed directly to validation and simulation logic.
Capability inventory: The skill lacks real execution sinks; logic is restricted to string matching and UUID generation.
Sanitization: The validate_code function implements a static blocklist check for keywords such as eval, exec, and subprocess to alert the agent to risky patterns.
[EXTERNAL_DOWNLOADS]: The skill documentation references the alibaba/OpenSandbox repository and assumes a local installation of the oss-agent-lab package. These are documented as legitimate dependencies for the simulation framework.

sandbox