Home > Core > Cognitive Modes

WaterfallMode

The "Plan-Ahead" engine: Implements a structured Plan -> Review -> Execute strategy for well-defined goals requiring human oversight.

Core Architecture Stateful Human-in-the-Loop
⚙️ WaterfallModeConfig.json 

{
  "type": "Waterfall",
  "planFile": "deploy_v1.json",
  "variables": {
    "target_env": "production",
    "version": "1.4.2"
  }
}
→
📊 Generated Plan Diagram
graph TD A[Read Config] --> B[Build Artifact] B --> C{Review Plan?} C -->|Approved| D[Deploy to production] D --> E[Verify Health]

Status: Awaiting User Approval (Discussable Phase)

The Waterfall Lifecycle

Unlike adaptive modes that "think as they go," WaterfallMode forces a separation between planning and action.

  1. Plan Generation: The agent analyzes the request and generates a full TaskBreakdown.
  2. Review (Discussable): If autoFix is false, the system pauses. The user can chat with the agent to refine the plan steps before any side effects occur.
  3. Variable Substitution: If using a planFile, placeholders like {{target_env}} are injected with runtime values.
  4. Execution: The TaskOrchestrator executes the validated graph sequentially.