The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides multiple tools that enable the execution of arbitrary code and shell commands within sandboxed environments.\n
The mcp__flow-nexus__sandbox_execute tool allows for the execution of arbitrary code in languages such as JavaScript and Python.\n
The mcp__flow-nexus__sandbox_configure tool includes a run_commands parameter, allowing the agent to execute arbitrary shell commands (e.g., npm run migrate).\n
The mcp__flow-nexus__sandbox_create tool supports a startup_script parameter for running setup commands at sandbox instantiation.\n- [CREDENTIALS_UNSAFE]: Tool parameters and documentation examples facilitate the handling of sensitive credentials.\n
Examples for mcp__flow-nexus__sandbox_create and mcp__flow-nexus__template_deploy demonstrate passing API_KEY, DATABASE_URL, and anthropic_key (using the sk- credential pattern) via environment variables.\n
Authentication management tools like mcp__flow-nexus__user_login and mcp__flow-nexus__user_update_password require the agent to process plaintext passwords and reset tokens.\n- [DATA_EXFILTRATION]: The skill's capabilities for reading and transmitting data create potential exfiltration paths.\n
The sandbox_execute tool allows code to perform network operations (e.g., fetch()), which could be used to send sensitive environment variables or file contents to external endpoints.\n
Tools such as mcp__flow-nexus__storage_upload and mcp__flow-nexus__sandbox_upload enable moving data from the agent's context to external storage buckets or remote sandboxes.\n- [EXTERNAL_DOWNLOADS]: Tools are provided to download and install external software from remote registries and repositories.\n
mcp__flow-nexus__sandbox_create and mcp__flow-nexus__sandbox_configure include an install_packages parameter for fetching dependencies from public package managers (npm, pip).\n
Examples in the advanced configuration section demonstrate the use of git clone to download source code from external repositories within the sandbox.\n- [PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection due to the processing of untrusted external data alongside high-privilege tools.\n
Ingestion points: The skill ingests untrusted code and data via mcp__flow-nexus__challenge_submit (solution_code), mcp__flow-nexus__app_store_publish_app (source_code), and mcp__flow-nexus__storage_upload (content).\n
Boundary markers: No explicit boundary markers or instructions are provided to the agent to isolate or ignore embedded instructions within the processed data.\n
Capability inventory: The skill provides extensive capabilities including arbitrary code execution, application deployment, and agent orchestration through the seraphina_chat interface.\n
Sanitization: No sanitization or validation logic for external content is described in the tool definitions.

flow-nexus-platform