skills/edwinhu/workflows/marimo/Gen Agent Trust Hub

marimo

Warn

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script marimo-serve/scripts/serve.py employs os.execv to replace the current Python process with the marimo editor binary. This process replacement relies on dynamic binary discovery via the system PATH using shutil.which.
  • [PERSISTENCE]: The documentation in marimo-serve/SKILL.md includes specific instructions for establishing persistent background processes on macOS using launchd configuration files (.plist) with the KeepAlive key enabled.
  • [COMMAND_EXECUTION]: When running in edit mode, the serve.py script invokes the marimo editor with the --no-token flag, effectively disabling authentication for the web-based kernel. While documented with warnings, this default configuration poses a risk of unauthorized access if the server is bound to an accessible network interface.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and manipulate Python-based notebook files. Since these files are parsed and their cell structure is analyzed (e.g., in scripts/get_cell_map.py), there is a potential surface for indirect prompt injection if a malicious notebook file is provided to the agent, potentially leading to unauthorized code execution or data exposure during the analysis or execution phase.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 5, 2026, 10:36 AM