Skip to content
Agentmart
Submit a tool

plan-mode

Use when a task needs a written execution plan with explicit steps, risks, verification, and user approval before implementation starts.

Skill by nexttylabs

IndexedMIT★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/nexttylabs-plan-mode

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when a task needs a written execution plan with explicit steps, risks, verification, and user approval before implementation starts.

Use cases: productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/plan-mode

What's inside

Skill name
plan-mode
Description
Use when a task needs a written execution plan with explicit steps, risks, verification, and user approval before implementation starts.
License
MIT
Instructions
121 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Plan Mode

A structured lifecycle for breaking complex work into verifiable steps, writing them to a plan document, and executing them autonomously.

## 1. Creating a Plan

### Gather context

Before writing anything, collect the information needed to plan well:

- Read relevant files, configs, and documentation.
- Identify constraints (time, risk, dependencies, available tools).
- Clarify ambiguous requirements with the user.

### Analyze

- Determine the minimal set of changes required.
- Identify independent vs. sequential steps.
- Flag anything that needs user approval before proceeding.

### Write the plan

Create a file at `plans/<slug>.md` using the format below. The slug should be a short, lowercase, hyphenated description (e.g., `add-retry-logic`).

### Emit checklist

After writing the plan file, print a numbered checklist summary so the user can review and approve it.

## 2. Plan Document Format

```markdown
# Plan: <title>

plan_id: <slug>
status: draft | approved | in_progress | blocked | completed | abandoned
current_focus: <step number or none>

## Goal

One or two sentences describing the desired outcome.

## Success Criteria

- [ ] Criterion 1
- [ ] Criterion 2

## Steps

### Step 1 — <short description>

- **Tools**: <which tools or commands this step uses>
- **Risk**: low | medium | high
- **Estimate**: <time or effort estimate>
- **Details**: What to do and what "done" looks like.

### Step 2 — ...

## Risks

- Risk description → mitigation strategy.

## Progress Log

| Step | Status      | Notes |
|------|-------------|-------|
| 1    | pending     |       |
| 2    | pending     |       |
```

## 3. Plan Rules

1. Each step must specify the **tools** it will use.
2. Each step must be **independently verifiable** — there should be a concrete way to confirm it succeeded.
3. Each step must include a **risk assessment** (low / medium / high).
4. Each step must include a **time estimate**.
5. A plan may have at most **20 steps**. If more are needed, split into multiple plans.
6. Steps should be ordered so that failures are caught early (high-risk or validation steps first when possible).

## 4. Approving and Executing

### Approve

The plan stays in `draft` status until the user explicitly approves it. Do not begin execution on a draft plan.

### Execute

1. Read the plan file to determine the current state.
2. Set plan status to `in_progress`.
3. Update `current_focus` to the step you are actively working on.
4. For each step marked `pending`:
   - Check `current_focus` — only work on one step at a time.
   - Execute the step using the specified tools.
   - Verify the result.
   - Update the Progress Log: set status to `done`, `failed`, or `skipped`, and add notes.
5. When all steps are resolved, set the plan status to `completed` and `current_focus` to `none`.

### Handle failures

- If a step fails, update its status to `failed` with a description of what went wrong.
- Attempt the step once more if the failure seems transient.
- If it fails again, set the plan status to `blocked` and report to the user. Do not skip ahead to later steps that depend on the failed one.

## 5. Checking Status

Re-read the plan file and print the Progress Log table. Highlight any steps that are `failed` or `blocked`.

## 6. Listing Plans

List all files in the `plans/` directory and print their `plan_id` and `status`.

## 7. Revising Plans

If requirements change mid-execution:

1. Set the plan status to `draft`.
2. Update the Goal, Steps, and Risks as needed.
3. Reset any not-yet-started steps to `pending`.
4. Set `current_focus` to `none`.
5. Get user approval again before resuming execution.

From nexttylabs/kubebar, MIT.

More from nexttylabs/kubebar

Signals

Stars
2
Forks
0
Open issues
0
Last commit (this folder)
Apr 18, 2026 (6 months ago)
Last release
Oct 4, 2026 (24 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from nexttylabs

Cite this page

plan-mode skill. Agent Mart. https://agentmart.dev/skills/nexttylabs-plan-mode (last updated 2026-10-24)