The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a legitimate interface for the agent to execute arbitrary Python code via the infsh/python-executor application. While this constitutes remote code execution, it is the primary intended function of the skill and is documented as operating within a sandboxed environment.
[EXTERNAL_DOWNLOADS]: The skill references external resources for installation and documentation, specifically fetching command-line interface instructions from the inference-sh GitHub repository. These downloads are functional requirements for the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a vulnerability surface where instructions embedded in the data being processed by the Python code could potentially influence the agent's behavior or the execution flow within the sandbox.
Ingestion points: The code parameter in the input schema (SKILL.md).
Boundary markers: Absent; the skill passes code directly to the execution engine.
Capability inventory: The execution environment includes network access (via requests, httpx), browser automation (via playwright, selenium), and file system writes to the outputs/ directory.
Sanitization: The skill relies on the platform's sandboxing mechanism rather than providing input-level sanitization.

python-executor