vrr/agent-zero
2
0
Fork 0
mirror of https://github.com/agent0ai/agent-zero.git synced 2026-05-19 16:31:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1
agent-zero/plugins/_a0_connector/prompts/agent.system.tool.code_execution_remote.md
Alessandro b48e31bead Forward remote exec reset flag 
Forward reset=true from code_execution_remote replacement commands to the connected CLI and document when to use it versus runtime=reset. This lets the CLI tear down stuck host sessions before running the next command.

Tests from /home/eclypso/a0/a0-connector: PYTHONPATH=src conda run -n a0 pytest tests/test_plugin_backend.py::test_code_execution_remote_forwards_reset_true_with_replacement_command -v; ./.venv/bin/python -m pytest tests/test_plugin_backend.py -k 'code_execution_remote or select_remote_exec or ws_connector_exec_result' -v. Mirrored to live container 07e0288dc04f and health check returned HTTP 200.
2026-05-15 00:28:10 +02:00

2.4 KiB
Raw Permalink Blame History

code_execution_remote tool

Runs shell-backed execution on the machine where a connected A0 CLI is running. Use this tool, not code_execution_tool, when the user asks for the connected local terminal, the A0 CLI host, their local machine, or explicitly says not to use Docker/server/container execution. For complex local project work, optionally load skill host-code-execution.

Availability and permissions are checked when the tool runs. If no CLI is connected, remote execution is disabled, or local access is not Read&Write for a mutating command, report that to the user instead of falling back to server-side execution.

Do not use this tool as a fallback for host-browser navigation/control. For "my browser", host browser, local browser/Chrome, or opening a URL in the host browser, use the browser tool. If Browser reports missing Chrome remote-debugging consent, tell the user to open chrome://inspect/#remote-debugging, enable "Allow remote debugging for this browser instance", run /browser host on, and retry.

Arguments

  • runtime: one of terminal, python, nodejs, output, reset
  • session: integer session id (default 0)
  • reset: optional boolean for terminal, python, or nodejs; when true, the CLI resets the session before running the supplied code

Runtime-specific fields:

  • terminal, python, nodejs: require code
  • reset: optional reason

Notes

  • Reuse session when continuing a workflow.
  • Use output to poll a running session and runtime=reset for a stuck session. Use reset: true on a new command when you need a clean session and want to run the replacement command immediately.
  • Paths and shell syntax are evaluated on the CLI host, not inside Agent Zero.
  • When the user gives a relative path like tmp/file.txt, keep it relative to the CLI host terminal. Do not prepend or cd to /a0/usr/workdir; that is the Agent Zero server/Docker workdir, not the connected local terminal folder.
  • If the current terminal folder matters, run pwd first or include pwd in the same command without changing directories.

Usage

{
  "thoughts": [
    "The user asked for the connected local terminal rather than Docker, so I should execute on the A0 CLI host."
  ],
  "headline": "Running command on connected local terminal",
  "tool_name": "code_execution_remote",
  "tool_args": {
    "runtime": "terminal",
    "session": 0,
    "code": "pwd"
  }
}