plan-writing – skill Viết lách & Nội dung cho agent AI

Plan Writing

> Source: obra/superpowers

Overview

This skill provides a framework for breaking down work into clear, actionable tasks with verification criteria.

Task Breakdown Principles

1. Small, Focused Tasks

• Each task should take 2-5 minutes
• One clear outcome per task
• Independently verifiable

2. Clear Verification

• How do you know it's done?
• What can you check/test?
• What's the expected output?

3. Logical Ordering

• Dependencies identified
• Parallel work where possible
• Critical path highlighted
• Verify each task as it is completed, then run a final acceptance check.

4. Dynamic Naming in Project Root

• Plan files are saved as {task-slug}.md in the PROJECT ROOT
• Name derived from task (e.g., "add auth" → auth-feature.md)
• Follow repository/user conventions when they specify another location; preserve an existing supplied plan.

Planning Principles (NOT Templates!)

> 🔴 NO fixed templates. Each plan is UNIQUE to the task.

Principle 1: Keep It SHORT

| ❌ Wrong | ✅ Right |

|----------|----------|

| 50 tasks with sub-sub-tasks | 5-10 clear tasks max |

| Every micro-step listed | Only actionable items |

| Verbose descriptions | One-line per task |

> Rule: If plan is longer than 1 page, it's too long. Simplify.

---

Principle 2: Be SPECIFIC, Not Generic

| ❌ Wrong | ✅ Right |

|----------|----------|

| "Set up project" | "Run npx create-next-app" |

| "Add authentication" | "Install next-auth, create /api/auth/[...nextauth].ts" |

| "Style the UI" | "Add Tailwind classes to Header.tsx" |

> Rule: Each task should have a clear, verifiable outcome.

---

Principle 3: Dynamic Content Based on Project Type

For NEW PROJECT:

• What tech stack? (decide first)
• What's the MVP? (minimal features)
• What's the file structure?

For FEATURE ADDITION:

• Which files are affected?
• What dependencies needed?
• How to verify it works?

For BUG FIX:

• What's the root cause?
• What file/line to change?
• How to test the fix?

---

Principle 4: Scripts Are Project-Specific

> 🔴 DO NOT copy-paste script commands. Choose based on project type.

| Project Type | Relevant Scripts |

|--------------|------------------|

| Frontend/React | ux_audit.py, accessibility_checker.py |

| Backend/API | api_validator.py, security_scan.py |

| Mobile | mobile_audit.py |

| Database | schema_validator.py |

| Full-stack | Mix of above based on what you touched |

Wrong: Adding all scripts to every plan

Right: Only existing, inspected scripts relevant to THIS task; the names above are illustrative, not bundled prerequisites

---

Principle 5: Verification is Simple

| ❌ Wrong | ✅ Right |

|----------|----------|

| "Verify the component works correctly" | "Run npm run dev, click button, see toast" |

| "Test the API" | "curl localhost:3000/api/users returns 200" |

| "Check styles" | "Open browser, verify dark mode toggle works" |

---

Plan Structure (Flexible, Not Fixed!)

# [Task Name]

## Goal
One sentence: What are we building/fixing?

## Tasks
- [ ] Task 1: [Specific action] → Verify: [How to check]
- [ ] Task 2: [Specific action] → Verify: [How to check]
- [ ] Task 3: [Specific action] → Verify: [How to check]

## Done When
- [ ] [Main success criteria]

> That's it. No phases, no sub-sections unless truly needed.

> Keep it minimal. Add complexity only when required.

Optional notes may record important considerations.

---

Best Practices (Quick Reference)

1. Start with goal - What are we building/fixing?

2. Max 10 tasks - If more, break into multiple plans

3. Each task verifiable - Clear "done" criteria

4. Project-specific - No copy-paste templates

5. Update as you go - Mark [x] when complete

---

When to Use

• New project from scratch
• Adding a feature
• Fixing a bug (if complex)
• Refactoring multiple files

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.