Skip to content
Back to skills

Plan

ASecurity

Activate the heavy planner→/orchestrate chain — write a structured plan file with Agent Dispatch Manifest, then invoke the /orchestrate skill to execute it in waves. Reserve for genuinely multi-file / multi-hour / multi-agent work (large refactors, migrations, features spanning many files). For single-session-sized work (one or a few files), use built-in plan mode (shift-tab) instead — not this skill.

  • 8 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added October 3, 2026
ai-agentscode-review

Security analysis

A100/100

Scanned October 3, 2026

npx -y skills add ek33450505/claude-agent-team --skill plan --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Plan?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Plan
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/ek33450505-plan/badge)](https://www.skillsdirectory.com/skills/ek33450505-plan)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: plan
description: Activate the heavy planner→/orchestrate chain — write a structured plan file with Agent Dispatch Manifest, then invoke the /orchestrate skill to execute it in waves. Reserve for genuinely multi-file / multi-hour / multi-agent work (large refactors, migrations, features spanning many files). For single-session-sized work (one or a few files), use built-in plan mode (shift-tab) instead — not this skill.
user-invocable: true
allowed-tools: [Write, Read, Glob, Grep, Agent]
---

# Plan Mode

This is the `/plan` skill — the heavy `planner`→`/orchestrate` chain, reserved for genuinely multi-file / multi-hour / multi-agent work. For single-session-sized tasks (one or a few files, finishable in one session), use built-in plan mode (shift-tab) with a single agent instead of this skill. You are entering plan mode to write a structured implementation plan.

## Step 1 — Write the plan file

Write a plan file under `~/.claude/plans/` using the Write tool:

- Filename: `~/.claude/plans/<YYYY-MM-DD>-<slug>.md` where slug is a short kebab-case description
- Include all of these sections:
  - Title and context
  - Fix strategy / implementation approach (one subsection per logical change)
  - Files to modify table
  - Implementation order
  - Verification steps
  - `## Agent Dispatch Manifest` — **mandatory** — a `json dispatch` code block with batches, agent types, prompts, and parallel flags

The ADM block must follow this schema:
```json dispatch
{
  "batches": [
    {
      "id": 1,
      "description": "short description",
      "parallel": true,
      "agents": [
        { "subagent_type": "agent-name", "prompt": "specific task prompt" }
      ]
    }
  ]
}
```

Use `"subagent_type": "main"` for tasks the orchestrator handles inline. Use named agents (e.g. `"code-reviewer"`, `"commit"`) for delegated tasks. Group independent tasks into parallel batches.

## Step 2 — Show plan summary and ask for approval

After writing the plan file, display a concise summary:
- Plan file path
- Number of batches and agents
- A table: Batch | Mode | What it does

Then ask the user: **"Execute this plan with /orchestrate? [yes/no]"**

Wait for explicit confirmation before proceeding.

## Step 3 — Execute via /orchestrate skill

Once the user confirms, invoke the `/orchestrate` skill with the plan file path:

```
/orchestrate [ABSOLUTE_PLAN_PATH]
```

The `/orchestrate` skill reads the plan's Agent Dispatch Manifest and executes all batches directly from the main session — presenting the batch queue, running the interrupt window, then dispatching agents in order.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.