agent-zero-bridge

Warn

Audited by Gen Agent Trust Hub on Apr 10, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The clawdbot_client.js script enables Agent Zero to invoke any tool configured on the Clawdbot Gateway via the /tools/invoke API endpoint. This provides a mechanism for a remote or delegated agent to execute powerful local commands without direct user oversight.
  • [DATA_EXFILTRATION]: The a0_api.js library and a0_client.js CLI tool support attaching arbitrary local files to messages sent to the Agent Zero API. The implementation uses fs.readFileSync on user-supplied paths, creating a risk of sensitive file exposure if the agent is manipulated into attaching configuration or credential files.
  • [PROMPT_INJECTION]: The task_breakdown.js script is vulnerable to indirect prompt injection. It interpolates untrusted task descriptions from the command line directly into a prompt template for Agent Zero.
  • Ingestion points: task_breakdown.js accepts raw task descriptions as CLI arguments.
  • Boundary markers: The input is wrapped in double quotes within the prompt, which is insufficient to prevent instruction override.
  • Capability inventory: The script can send messages and attachments to the Agent Zero API and has write access to the notebook/tasks/projects/ directory.
  • Sanitization: No sanitization or validation of the input task description is performed before interpolation.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 10, 2026, 01:34 PM