Writing Implementation Plans
Purpose
Create comprehensive plans assuming the implementer has zero context.
"I'm using the Writing Plans skill to create the implementation plan."
Save to: docs/plans/YYYY-MM-DD-<feature-name>.md
Plan Header Template
# [Feature Name] Implementation Plan
**Goal:** [One sentence]
**Architecture:** [2-3 sentences]
**Tech Stack:** [Key technologies]
**Estimated Tasks:** [count]
---Task Granularity
Each step = ONE action (2-5 minutes):
- "Write the failing test" ← step
- "Run it to confirm failure" ← step
- "Implement minimal code" ← step
- "Run tests, confirm pass" ← step
- "Commit" ← step
NOT: "Write tests and implement feature" ← too big
Task Structure
### Task N: [Component Name]
**Files:**
- Create: `exact/path/to/file.ts`
- Modify: `exact/path/to/existing.ts:lines`
- Test: `tests/path/to/test.ts`
**Step 1: Write failing test**
\`\`\`typescript
test('should do X', () => {
expect(thing()).toBe(expected);
});
\`\`\`
**Step 2: Verify failure**
Run: `npm test -- --grep "should do X"`
Expected: FAIL - "thing is not defined"
**Step 3: Implement**
\`\`\`typescript
export function thing() {
return expected;
}
\`\`\`
**Step 4: Verify pass**
Run: `npm test -- --grep "should do X"`
Expected: PASS
**Step 5: Commit**
\`\`\`bash
git add -A && git commit -m "feat: add thing"
\`\`\`Plan Principles
| Principle | Meaning | |-----------|---------| | DRY | Don't repeat yourself | | YAGNI | Don't build what's not needed | | TDD | Test first, always | | Small commits | One logical change per commit |
What to Include
- ✅ Exact file paths
- ✅ Complete code snippets (not "add validation")
- ✅ Exact commands with expected output
- ✅ Dependencies to install
- ✅ Environment setup needed
Handoff
After saving plan:
**Plan saved to `docs/plans/[filename].md`**
Ready to execute. Options:
1. **Execute now** - I'll work through tasks with you
2. **Subagent execution** - Fresh agent per task with review gates
3. **Later** - Plan saved for future sessionIf executing → Use executing-plans skill