agent-manager-skill

Warn

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation instructs the user to clone a repository from https://github.com/fractalmind-ai/agent-manager-skill.git. This source is external and does not belong to a recognized trusted vendor or the skill author.
  • [REMOTE_CODE_EXECUTION]: The skill directs the user to execute local Python scripts (agent-manager/scripts/main.py) which are downloaded from the external GitHub repository, enabling the execution of arbitrary code.
  • [COMMAND_EXECUTION]: The skill uses python3 and tmux to manage agent sessions and system processes, which involves executing arbitrary shell commands on the host.
  • [PROMPT_INJECTION]: The assign command creates a surface for indirect prompt injection by interpolating untrusted instructions into agent sessions via a heredoc block. Ingestion points: assign command in SKILL.md. Boundary markers: Heredoc EOF is present. Capability inventory: Subprocess execution via python3 and tmux session management. Sanitization: No validation or sanitization of the input instructions is mentioned.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 21, 2026, 03:22 PM
Security Audit — agent-trust-hub — agent-manager-skill