| name | plan |
| description | Agent Skill for creating an implementation plan. Use when the user asks to plan, formalize, or save implementation work, or turn defined TODOs into sequenced, verifiable steps. |
Plan
Create an implementation plan that contains clear, self-contained, and verifiable steps.
Prerequisites
The user or you AI Agent should already have a clear context or clear direction in mind. If not, use brainstorming skill first
Format of the plan
The plan should be written in markdown.
Beginning format
The beginning format should be:
# [Feature Name] Implementation Plan
## Goal
[No more than 3 sentences describing what this changes about]
## Context
[Any relevant context, background, scope, and success criteria]
[If there is a source-of-truth, such as the spec, issue, or a link, include it here]
## Approach
[2-3 sentences about the approach]
You can add more sections as needed.
Steps Section
Begin with ## Steps section like following:
## Steps
When executing the plan:
- Mark `[ ]` boxes as completed `[x]` when item is completed
- After the imeplementation and verification of each step, spawn a subagent to do the code review, and fix any valuable feedbacks
- If possible, reuse the same review subagent between steps
- Before moving to the next step, commit the changes
Step format
For each step, placed inside the ## Steps section with following structure:
### [ ] Step N: [Step Name]
[One sentence describing what this step accomplishes]
[Optionally, a list of dependent steps in comma-separated one line list, beginnig with `**Depends on:**`]
#### Implementation
[Concise, concrete, and clear instructions for implementation]
#### Verification
[How to verify the implementation]
#### Notes
[Optional: any notes about this step, this section can be omitted if none]
Reviewing the plan
After writing the plan, spawn a subagent with a fresh context window to review the plan. Address any valuable feedbacks from the subagent.
Run up to 3 review rounds, stopping early when a reviewer finds no actionable problems.
If the review agent spots a problem that needs more clarification or more planning work, or anything that invalidate the plan, surface the problem to the user for guidance.
Writing the plan
By default, the plan should be saved in plans/plan-<slug>-YYYYMMDD.md.
However, project may have a different plans folder and/or different naming convention.
Always scan for project-specific plans folder and naming convention, and follow it if found.
Notes
About the overall procedure
- The plan file must be standalone and self-contained, which means, any one (including a new AI Agent) who is new to this project, can read the plan and start the implementation without guess work
- Make use of mermaid diagrams or ascii diagrams when they can clarify architecture, flow, state transition,compnents relationship, or any other explaination
About each step
- When splitting the work into the step, split based on the size, and the high-cohesion low-coupling principle.
- The size of each step should be a size reviewable by a human.
- Each step should be meaningful, self-contained, and self-verifiable.
- Include enough implementation detail to avoid guesswork, but do not turn the plan into a full code dump
- Use snippets, interfaces, or pseudocode if they can help reduce ambiguity.
