agent-manager-skill
Fail
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs users to clone source code from a repository not associated with the author or trusted vendors:
https://github.com/fractalmind-ai/agent-manager-skill.git. - [REMOTE_CODE_EXECUTION]: The skill executes Python scripts (
main.py) directly from the downloaded repository, allowing for the execution of unverified third-party code on the user's environment. - [PERSISTENCE]: The skill description mentions "cron-friendly scheduling," indicating it manages system cron jobs to maintain recurring execution of agent tasks, a technique used for persistence.
- [COMMAND_EXECUTION]: Functionality relies on shell commands for Python script execution and managing parallel background processes via
tmuxsessions. - [INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: The
assigncommand accepts workflows via stdin (EOF). 2. Boundary markers: None identified in the provided documentation. 3. Capability inventory: Python execution, tmux management, and crontab modification. 4. Sanitization: No evidence of validation or filtering for task inputs is present.
Recommendations
- AI detected serious security threats
Audit Metadata