Determine Next Steps

This skill allows the agent to determine and structure the immediate next steps before executing anything. It ensures actions are ordered by dependency, gated by the current action_gate mode, and flagged for risks before execution begins.

Steps

1. Read current state fresh. Apply Zero-Drift Audit Law — read .db/.system.board.yaml and any relevant session/goal files live before planning.
2. Clarify the goal. State the objective in one sentence. If the goal is vague, apply analyze-context first to sharpen it.
3. Check the action_gate. Read the action_gate from .db/.system.board.yaml:
   • PLANNING → produce the plan only, do NOT execute.
   • EXECUTION → produce the plan AND proceed to execute step 1 immediately.
4. Map dependencies. List what must be true or done BEFORE each step can start. Draw a simple dependency order.
5. Order actions by dependency. Steps that unblock others come first. Parallelizable steps are noted explicitly.
6. Identify checkpoints. Define verification points — moments where the agent must confirm state before proceeding.
7. Flag risks. Note any step that is irreversible, destructive, or uncertain. These become hard gates requiring confirmation.
8. Output the structured plan.

Output Format

action_plan:
  goal: "<single sentence goal>"
  action_gate: "PLANNING | EXECUTION"
  steps:
    - id: 1
      action: "<what to do>"
      depends_on: []
      reversible: true | false
      checkpoint: false
    - id: 2
      action: "<what to do>"
      depends_on: [1]
      reversible: false
      checkpoint: true   # Must verify before proceeding
  risk_flags:
    - step: 2
      risk: "<description>"
      mitigation: "<what to do if it fails>"

Notes

• Never skip dependency mapping — executing steps out of order causes cascading failures.
• In PLANNING mode, this skill is the final output. Do not proceed to execution.
• In EXECUTION mode, return to assess-quality after each checkpoint step.