The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were identified. The skill follows established practices for integrating with the Acontext platform for storage and data processing.
[COMMAND_EXECUTION]: The skill facilitates command execution through tools like bash_execution_sandbox and analyze_data. These tools are designed to operate within the restricted environment of an Acontext Sandbox, which is the intended functionality for providing code interpretation capabilities to the agent.
[EXTERNAL_DOWNLOADS]: During sandbox initialization for data analysis, the skill downloads and installs the seaborn and pandas Python packages. These downloads are performed from the official PyPI registry, which is an established and trusted package repository.
[REMOTE_CODE_EXECUTION]: The executeSandboxTool function allows for the execution of Python scripts generated by the LLM. This execution is confined to a clean sandbox container created for each session, mitigating the risks typically associated with runtime code execution.
[PROMPT_INJECTION]: The skill processes untrusted input from conversation history and user-stored files (Category 8), which creates a potential surface for indirect prompt injection. This data enters the agent's context through functions such as loadMessages and readFile. While the code lacks explicit boundary markers for these inputs, the risk is inherent to the chatbot's primary function of processing and responding to user-provided information, and the available actions are limited to sandboxed operations.

acontext-chatbot-integration