# Plan Task

> Breaks the next task from a plan into an implementable task file. Use when decomposing a multi-step plan into ordered work units.

- Skill: `haowjy/plan-task` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add haowjy/plan-task`
- Raw SKILL.md: https://api.skillmd.com/api/skills/haowjy/plan-task/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: haowjy (https://skillmd.com/u/haowjy)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/haowjy/plan-task

---


# Plan Task

Create the next implementable task from a plan.

## When Invoked

### Step 1: Read Inputs

Use these prompt variables:
- `{{PLAN_FILE}}` — source plan to read

Resolve output directory:
- Always use a path next to the plan: `<dir-of-PLAN_FILE>/tasks`.
- Create the output directory if it does not exist.

### Step 2: Read Progress

Gather context on what is already completed:

1. Read `{{PLAN_FILE}}` and inspect status/phase tracking sections.
2. If `<resolved-tasks-dir>/progress.md` exists, read it for prior completed tasks.
3. If `<resolved-tasks-dir>/task.md` exists, use it as context for continuity.

### Step 3: Determine Completion

If all phases/steps in the plan are complete:
- Write **only** the text `ALL_DONE` to `<resolved-tasks-dir>/task.md`.
- Stop here.

### Step 4: Create Task File

Determine the next logical task from the plan.

Write `<resolved-tasks-dir>/task.md`. Include whatever structure makes sense, but the task file should give an implementing agent everything it needs:

- **Context**: why this task is next
- **Scope**: what to implement — files, functions, integration points
- **Acceptance criteria**: how to verify it's done
- **Constraints**: architectural limits, gotchas

A good task is **self-contained**: the codebase must be in a working state when done. Size is secondary — focus on logical completeness, not line count.