# Determine Next Steps

> Structures immediate next steps into a dependency-ordered plan, gated by the current action_gate mode, with checkpoints and risk flags before execution.

- Skill: `auto-skiller/determine-next-steps` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add auto-skiller/determine-next-steps`
- Raw SKILL.md: https://api.skillmd.com/api/skills/auto-skiller/determine-next-steps/raw
- Safety review: PASS (external: skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity, Task Management
- Tags: Action Gate, Dependency Mapping, Planning, Risk Assessment, Task Planning
- Author: Auto-Skiller (https://skillmd.com/u/auto-skiller)
- Updated: 2026-08-22
- Page: https://skillmd.com/skills/auto-skiller/determine-next-steps

---


# 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

```yaml
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.