Hanzo Dev

Auto Drive

Fully automated multi-turn task execution in Hanzo Dev.

Auto Drive provides fully automated, multi-turn task execution with planning, agent orchestration, and review gates.

Start Points

  • TUI: /auto <goal> — if you omit the goal and there is recent history, Hanzo Dev proposes one for you
  • CLI: dev exec --auto "<goal>" or dev exec "/auto <goal>" — a goal is required headless
  • Precondition: Full Auto mode must be selected in the TUI; otherwise Auto Drive will not start

How It Runs

Each turn, Auto Drive:

  1. Drafts a plan
  2. Prepares commands
  3. Optionally assigns agents
  4. Waits for your confirmation (or the countdown) before running

The transcript is kept in memory and compacted automatically.

Agents

Auto Drive can spawn helper agents during a turn. Toggle with agents_enabled in Auto Drive settings. Outside a git repo, agents are forced to read-only.

Observer

A lightweight watchdog reviews the run every auto_drive_observer_cadence turns (default 5). Set to 0 to disable.

Continue Modes

ModeBehavior
immediateProceed without waiting
ten-seconds10-second countdown (default)
sixty-seconds60-second countdown
manualPause after each prepared prompt

In countdown modes, press Enter to proceed early, Esc to reopen the draft, or 0 to auto-submit.

Review & QA

  • review_enabled (default true) inserts review gates
  • qa_automation_enabled and cross_check_enabled allow diagnostics and cross-checks
  • auto_resolve_review_attempts limits auto-resolve iterations (default 5)

Configuration

# Top-level keys
auto_drive_use_chat_model = false
auto_drive_observer_cadence = 5

[auto_drive]
review_enabled = true
agents_enabled = true
qa_automation_enabled = true
cross_check_enabled = true
observer_enabled = true
coordinator_routing = true
continue_mode = "ten-seconds"
model = "gpt-5.1"
model_reasoning_effort = "high"
auto_resolve_review_attempts = 5

Tips

  • Stay in the TUI for countdowns and visual status; use dev exec --auto for CI
  • If Auto Drive stops because it couldn't derive a goal, rerun /auto <goal> with specific instructions
  • Turn off agents in /auto settings for single-model runs

On this page