persistent-localhost

Warn

Audited by Gen Agent Trust Hub on Sep 13, 2026

Risk Level: MEDIUMPERSISTENCECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill's primary function is to create and manage macOS LaunchAgents by writing property list files to ~/Library/LaunchAgents/. These services are configured to run automatically at load and automatically restart on failure, allowing the agent to establish persistent execution that outlives the current session.
  • [COMMAND_EXECUTION]: The scripts/persistent-localhost.sh script facilitates the execution of arbitrary shell commands. The string passed to the --cmd argument is executed via /bin/bash -c within the persistent service context, providing a mechanism for long-running process execution.
  • [DYNAMIC_EXECUTION]: The tool dynamically generates system configuration files (XML-based .plist files) at runtime based on user-supplied parameters. It then loads these configurations into the system's service manager using launchctl bootstrap.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by allowing external command strings to be incorporated into persistent system-level services.
  • Ingestion points: The --cmd and --env arguments in the scripts/persistent-localhost.sh script, which are populated from agent instructions.
  • Boundary markers: No boundary markers or "ignore instructions" warnings are present to delimit the command string within the plist configuration.
  • Capability inventory: The skill possesses capabilities for persistent process execution, file system writes in the user's library, and process termination (kill) for port management.
  • Sanitization: The script includes a basic XML escaping function to ensure valid plist syntax, but it does not perform safety validation or sanitization on the shell command logic itself.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 13, 2026, 03:05 PM
Security Audit — agent-trust-hub — persistent-localhost