dev-swarm-headless-ai-agents

Warn

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill provides explicit instructions and command-line flags (--dangerously-skip-permissions, --yolo, --ask-for-approval never) designed to bypass the safety and permission guardrails of AI agents.
  • [COMMAND_EXECUTION]: The documentation encourages the execution of AI agents in a mode that automatically approves all system actions, including file writes and shell command execution, without user intervention.
  • [REMOTE_CODE_EXECUTION]: The Python code examples provided in the reference files use the subprocess module to execute system commands with unsanitized user-provided prompts, presenting a potential command injection surface if integrated into other applications.
  • [PROMPT_INJECTION]: The skill demonstrates a vulnerability surface for indirect prompt injection (Category 8).
  • Ingestion points: The skill instructs users to pipe local files (e.g., cat main.py) and version control data (e.g., git diff) directly into AI agents.
  • Boundary markers: No delimiters or protective instructions are used to separate untrusted data from the agent's instructions.
  • Capability inventory: Agents are executed with flags that permit automated file writes and command execution without confirmation.
  • Sanitization: No sanitization or validation of the ingested data is recommended before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 13, 2026, 08:28 PM