The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides the mcp__flow-nexus__sandbox_execute tool, which allows the agent to execute arbitrary JavaScript or code provided as a string within a remote sandbox environment.\n- [COMMAND_EXECUTION]: The tools mcp__flow-nexus__sandbox_create and mcp__flow-nexus__sandbox_configure allow for the execution of arbitrary shell commands via the startup_script and run_commands parameters respectively.\n- [EXTERNAL_DOWNLOADS]: Sandbox management tools permit the installation of arbitrary third-party packages (e.g., via install_packages) and the execution of commands that can download external content, such as git clone or apt-get install shown in platform examples.\n- [DATA_EXFILTRATION]: The skill includes tools for uploading and managing files in cloud storage (mcp__flow-nexus__storage_upload) and transferring files into sandbox environments (mcp__flow-nexus__sandbox_upload), which could be leveraged to move sensitive data out of the agent's local context or between environments.\n- [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by ingesting untrusted data such as application source code, challenge solutions, and chat messages which are then processed or executed by platform tools or the Seraphina AI assistant.\n
Ingestion points: Untrusted content enters via mcp__flow-nexus__app_store_publish_app (source_code), mcp__flow-nexus__challenge_submit (solution_code), and mcp__flow-nexus__seraphina_chat (message).\n
Boundary markers: None identified in the tool instructions to delimit user-provided code from system instructions.\n
Capability inventory: The skill possesses high-privilege capabilities including remote code execution, shell command execution, and cloud file storage access.\n
Sanitization: There is no evidence of sanitization or validation of ingested code or messages before they are processed by the platform tools.

flow-nexus-platform