adaptive-planning

name	adaptive-planning
description	Use when an approved design needs an implementation plan, and plan depth should adapt to risk and scope.

Skill: adaptive-planning

Overview

Turn an approved design into an execution plan that is short by default and deep when risk warrants. This skill produces and gates the plan; it does not implement.

Workflow

1. Read the approved design and relevant repository context (touched files, existing conventions, prior failures if known).
2. Classify scope: trivial / standard / high-risk (see Depth Selection Rules below).
3. Choose plan depth:
   • Short plan (default): 5–10 concrete bullets in execution order.
   • Deep plan: task-by-task checklist with exact files, test cases, verification checkpoints, and commit slices.
4. Present the plan to the user.
5. Require explicit user approval before implementation starts. Do not proceed to code without it.

Depth Selection Rules

Auto-switch to deep when any of the following are true:

• 3 or more files will be touched
• Migration or data model changes are involved
• Auth, security, or permission logic is affected
• External API contract is being changed or introduced
• Behavior is unknown, flaky, or prior implementation attempts failed

Overrides

• If the user requests a "deep plan" explicitly, force deep regardless of scope.
• If the user requests a short plan, keep it short unless critical risk (security, data integrity, external contract) makes a shallow plan unsafe.

Guardrails

• If you have cesium tools available, use those for generating artifacts.
• Keep plans minimal and actionable. Avoid speculative refactors or changes outside stated scope.
• Preserve existing project constraints, coding conventions, and guardrails found in CLAUDE.md, AGENTS.md, or GEMINI.md.
• Do not begin implementation from this skill. Stop after plan approval and hand off to the implementer.