| name | df-implementation-planning |
| description | Write a concrete, small-step implementation plan before coding. Use for multi-step features, bug fixes, refactors, DDD-to-TDD handoffs, risky behavior changes, or any engineering task that needs files, commands, expected RED/GREEN results, verification steps, and completion criteria before implementation. |
Implementation Planning
Use this skill to turn a requirement, confirmed DDD handoff, Glue Coding pattern selection, selected style pack, bug investigation result, or refactor goal into a plan that another agent can execute.
Do not edit production code while using this skill.
Planning Workflow
- Restate the goal and scope.
- List constraints:
- behaviors that must not change
- public contracts
- data, persistence, ordering, pagination, security, or side-effect risks
- selected style pack, golden examples, style-specific anti-patterns, and review checklist items to preserve
- selected glue target pattern, local conventions, legacy behavior evidence, anti-patterns to remove, and project material that must be preserved
- user-owned worktree changes to avoid
- Identify behavior slices.
- For each task, specify:
- objective
- files or modules likely involved
- test or verification command
- expected RED, GREEN, or unchanged result
- completion standard
- Keep each task small enough to complete and verify independently.
- Mark steps that require
df-tdd-skill, df-spring-web-boundaries, or df-systematic-debugging.
- For glue-style work, include the selected style pack when one exists, the selected target pattern, and the exact delta each task is allowed to change.
- For refactor glue work, separate characterization, target-pattern migration, and cleanup steps. Do not plan to copy legacy structure unless it is explicitly classified as the target pattern.
- State whether user confirmation is required before execution.
Step Size
Prefer 2-5 minute execution steps. Split a step when it combines:
- creating tests and implementing production code
- unrelated behavior slices
- multiple modules with separate risks
- refactor and behavior change
- debugging and fixing
- code changes and commit/PR work
Output Format
Use templates/implementation-plan.md when a file artifact is useful. In chat, use the same structure:
# <Name> Implementation Plan
## 目标
## 约束
## 行为切片
## 任务列表
1. 写失败测试:<behavior>
- 文件:
- 命令:
- 预期 RED:
- 完成标准:
2. 最小实现:<behavior>
- 文件:
- 命令:
- 预期 GREEN:
- 完成标准:
3. 重构:<design cleanup>
- 文件:
- 命令:
- 必须保持:
- 完成标准:
## 验证矩阵
## 需要用户确认
Non-Negotiable Rules
- Do not write production code in this phase.
- Do not produce vague steps such as "implement feature" or "run tests".
- Do not omit commands when the project has discoverable test commands.
- Do not plan a fix before a reproducible failure and root-cause evidence exists for unclear bugs.
- Do not mix behavior change and broad cleanup in the same step.
- Do not invent new structure for glue-style work when a selected style pack or local target pattern must be preserved.
- Do not plan refactors that treat legacy code or anti-patterns as the target pattern without explicit justification.
- Do not assume approval for risky scope expansion; call it out.