duoduo-runtime-admin

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to perform administrative tasks using standard command-line tools such as duoduo, git, npm, python, curl, and file manipulation utilities (find, mv). These are used for daemon diagnostics, configuration updates, and log management.\n- [EXTERNAL_DOWNLOADS]: The skill includes a procedure to refresh subconscious partition prompts by fetching content from the vendor's official GitHub repository (openduo/duoduo). This is a legitimate update mechanism for the agent's internal prompts.\n- [DYNAMIC_EXECUTION]: Provides a playbook for diagnostics that involves generating a temporary payload using a Python heredoc. This is used to create specific JSON request bodies for probing model context windows.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external endpoints and model profiles. It mitigates risks by requiring human confirmation of diffs before overwriting partition prompts and using a dedicated script with validation for environment variable updates.\n- [SAFE]: The skill demonstrates strong security posture by:\n
  • Implementing a 'custody rule' for API keys, ensuring they are never emitted in logs or chat history.\n
  • Using secure input methods (stdin, env vars, or 0600-permission files) for credentials to avoid exposure in process lists.\n
  • Masking sensitive values in all display and configuration paths.\n
  • Enforcing a diff-before-overwrite discipline for subconscious refreshes to protect local modifications.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 12:53 AM
Security Audit — agent-trust-hub — duoduo-runtime-admin