integrate-anything

Pass

Audited by Gen Agent Trust Hub on Jun 28, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill instructs the agent to follow agentInstructions provided by the Membrane service "verbatim" if they are present in the connection details. This creates a surface for indirect prompt injection, where malicious instructions from an external integration or compromised service could influence the agent's actions.\n
  • Ingestion points: clientAction.agentInstructions field from CLI command outputs in SKILL.md.\n
  • Boundary markers: Absent; instructions are explicitly to be followed "verbatim".\n
  • Capability inventory: Shell command execution via CLI, authenticated network requests, and dynamic JavaScript execution.\n
  • Sanitization: Absent.\n- [COMMAND_EXECUTION]: The skill utilizes the Membrane CLI (npx @membranehq/cli) for all core functions, including session management and action execution, involving shell commands with potentially dynamic arguments.\n- [EXTERNAL_DOWNLOADS]: The skill relies on downloading and installing the @membranehq/cli package from the npm registry using npx or global installation.\n- [REMOTE_CODE_EXECUTION]: The act --code functionality allows the agent to generate and execute JavaScript code on the Membrane platform. While intended for mapping and logic, it represents a pattern of dynamic execution of agent-generated code.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 28, 2026, 05:51 PM
Security Audit — agent-trust-hub — integrate-anything