The Agent Skills Directory

[COMMAND_EXECUTION]: The management script scripts/neuralos.mjs and the Windows installation script service/install-windows-service.ps1 execute system-level commands to manage processes and register services. This includes the use of sudo for Linux setup and elevated PowerShell with execution policy bypasses for Windows Task Scheduler registration.
[COMMAND_EXECUTION]: The skill establishes system-wide persistence by creating and enabling system-level service units (systemd .service, launchd .plist) and scheduled tasks to ensure the backend daemon remains active across system reboots.
[EXTERNAL_DOWNLOADS]: Setup procedures trigger package downloads from the npm registry for the neuralos and rterm-backend packages, as well as the wigolo tool used by the web-intelligence plugin.
[REMOTE_CODE_EXECUTION]: The skill dynamically interprets and executes dagu YAML workflows by compiling them into playbooks that run across orchestrated terminal sessions.
[PROMPT_INJECTION]: The skill processes untrusted data via external YAML configurations and workflows, creating a surface for indirect prompt injection.
Ingestion points: Processes external dagu YAML workflows via agent:startTask and YAML-based policies in the data directory.
Boundary markers: Instructions rely on the agent's ability to distinguish data from instructions within the task context; boundary markers are not explicitly enforced for these inputs.
Capability inventory: The skill manages a backend with full terminal access (SSH, WinRM, local PTY) and system administration capabilities.
Sanitization: The backend implements internal validation and rollback mechanisms for playbooks, though static analysis of input data content is not performed.

neuralos