The Agent Skills Directory

Remote Code Execution (HIGH): The installation script curl -fsSL https://cli.inference.sh | sh represents an untrusted remote code execution pattern. This executes unverified code from an external domain not recognized as a trusted source, creating a high-risk vector for system compromise.\n- Dynamic Execution (MEDIUM): The core functionality of the skill is to execute arbitrary Python code on remote infrastructure via the infsh CLI. While this is the intended primary purpose, providing an unconstrained execution environment for code strings poses a risk if input is manipulated by malicious actors.\n- Indirect Prompt Injection (LOW): The skill supports web scraping and external data processing, which creates an attack surface for indirect prompt injection.\n
Ingestion points: Data is ingested via requests.get and other network libraries from arbitrary, untrusted URLs.\n
Boundary markers: No delimiters or safety warnings are present in the provided code examples to separate ingested content from system instructions.\n
Capability inventory: The skill allows network access, file system writes to the outputs/ directory, and full Python execution logic.\n
Sanitization: No sanitization or validation of external content is demonstrated before it is processed by the agent.

python-executor