Task Planning
Break approved work into bite-sized, unambiguous tasks.
Announce: "I'm using task-planning to create detailed implementation tasks."
Iron Law
code
EACH TASK MUST BE COMPLETABLE IN 2-5 MINUTES WITH ZERO CONTEXT
Assume the implementing agent:
- •Has never seen this codebase
- •Doesn't know your reasoning
- •Will follow instructions literally
- •Needs exact file paths and code
Task Structure
markdown
### Task N.M: [Specific action]
**Files:**
- Create: `exact/path/to/new-file.ts`
- Modify: `exact/path/to/existing.ts:42-58`
- Test: `exact/path/to/test.spec.ts`
**Step 1: Write failing test**
```typescript
// exact code to add
test('specific behavior', () => {
const result = functionName(input);
expect(result).toBe(expected);
});
Step 2: Verify test fails
Run: bun test path/to/test.spec.ts
Expected: FAIL with "functionName is not defined"
Step 3: Implement
typescript
// exact code to add
export function functionName(input: Type): ReturnType {
return expected;
}
Step 4: Verify test passes
Run: bun test path/to/test.spec.ts
Expected: PASS
Step 5: Commit
bash
git add path/to/files git commit -m "feat(scope): specific change"
code
## Task Categories Group tasks by domain: ```markdown ## 1. Database Schema - [ ] 1.1 Add column X to table Y - [ ] 1.2 Create RLS policy for Z ## 2. Database Functions - [ ] 2.1 Create function A - [ ] 2.2 Modify function B ## 3. Frontend Components - [ ] 3.1 Create component X - [ ] 3.2 Update store Y ## 4. Integration - [ ] 4.1 Wire frontend to new function - [ ] 4.2 Add E2E test
Task Ordering
- •Database first - Schema, then functions
- •Tests alongside - Each implementation task includes test
- •Frontend after backend - Don't build UI for non-existent data
- •Integration last - Wire things together
Verification Commands
Every task includes verification:
| Layer | Command | Expected |
|---|---|---|
| Database | supabase test db | All tests pass |
| Frontend unit | bun test [file] | Specific test passes |
| E2E | bun run test:e2e | Flow works |
| Types | bun run type-check | No errors |
| Lint | bun run lint | No errors |
REQUIRED SUB-SKILL
Tasks ready → Load openspec-apply to begin implementation
Each code task → Load test-tdd for the implementation
Red Flags - STOP
If you catch yourself:
- •Writing tasks like "implement the feature"
- •Missing exact file paths
- •Missing verification steps
- •Tasks longer than 5 minutes
- •Skipping test steps
STOP. Break it down further.
Common Rationalizations
| Excuse | Reality |
|---|---|
| "The agent will figure it out" | They won't. Be explicit. |
| "Too much detail is patronizing" | Too little detail causes errors. |
| "This task is obvious" | Obvious to you ≠ obvious to fresh agent. |
| "I'll explain if they get stuck" | They can't ask you. Write it down. |