Skip to content
Goal Mode Configuration

Goal Mode Configuration

Configure Goal mode (Autopilot) behavior in .pando.toml.

Basic Configuration

[Goal]
# Maximum iterations before timeout (0 = default 20)
MaxIterations = 20

# Maximum duration (Go duration string)
MaxDuration = '1h'

# Consecutive no-progress iterations before stalled
StallIterations = 3

# Auto-approve all tool calls during goal mode
AutoApprove = true

# Patterns to block in goal mode (regex)
DangerousPatterns = []

Goal States

Goals progress through these states:

StateDescription
runningGoal is actively being pursued
completedObjective achieved
failedObjective cannot be achieved
blockedGoal is blocked by external factors
cancelledUser cancelled the goal
timeoutMax iterations or duration exceeded
stalledNo progress for N iterations

Non-Interactive Goal Mode

From the command line:

pando --goal "Fix all failing tests"
pando --goal "Refactor auth module" --model copilot.gpt-5.4
pando --goal "Add comprehensive error handling" --quiet

The CLI returns a structured result:

{
  "session_id": "...",
  "objective": "Fix all failing tests",
  "status": "completed",
  "iteration": 5,
  "response": "All 12 tests now pass",
  "progress": "Fixed authentication, database, and API tests",
  "next_step": null,
  "blocked_reason": null
}

Slash Commands

CommandDescription
/goal <objective>Start goal mode
/autopilot <objective>Alias for /goal
/goal-statusShow current goal status
/goal-cancelCancel running goal

Evaluation

The HeuristicGoalEvaluator analyzes each iteration for:

  • Completion signals: Tests passing, build success, explicit completion statements
  • Blocking signals: Error messages, missing dependencies, unresolvable issues
  • Stalling: No meaningful progress for N consecutive iterations

Safety

  • AutoApprove = true skips permission prompts during goal mode
  • DangerousPatterns blocks commands matching regex patterns
  • Ctrl+C cancels the running goal (instead of exiting Pando)
  • Max iterations and duration provide hard limits
Goal mode is powerful for autonomous tasks but should be used with caution. Start with simple objectives and gradually increase complexity as you trust the system.