ai-coding-coach

Files (1)

• SKILL.md

SKILL.md (first lines)

# AI Coding Coach

Implements code task-by-task with quality gates at every step.

## Modes (Bab 27.3 Tahap 4–5)

This skill has two modes. The active mode is declared by HANDOFF-PROMPT.MD
or by the user; if undeclared, start in MODE DRY RUN.

**MODE DRY RUN (mandatory first — never write code in this mode):**
1. Run the bootstrap command block from HANDOFF-PROMPT.MD (init script /
   dev-server + sanity check).
2. Read the pack + AGENTS.MD + STATE.MD (+ DECISIONS.MD, CONSTRAINTS.md).
3. Report the plan: tasks in order, files to create/modify, dependencies,
   risks, and the DoD commands that will prove each task.
4. Write the report to `planning/sprints/sprint-XXX/DRYRUN-REPORT.MD`
   (folder is the source of truth — the report must survive the session).
5. STOP and WAIT for explicit human approval. No approval = no build.

**MODE BUILD (only after approval):**
- Execute task-by-task via Steps 1–7 below, honoring the approved report.
- If reality diverges from the approved dry run (extra files, new
  dependency, scope change): STOP, report the divergence, get re-approval.
  Do not silently amend an approved plan.

## Input

Requires a confirmed module plan (output of `/module-planner`).
If no plan exists, stop and tell the user to run `/module-planner` first.

## Workflow

### Step 1: Pick the Next Task

Read the module plan. Find the first uncompleted task in execution order.
Display it:

```
📋 Current Task: [Task ID] - [Task Name]
📦 Module: [Module Name]
⏱ Estimated: [Time]
🔗 Depends on: [Previous tasks]
✅ Done when: [Clear acceptance criteria]
```

### Step 2: Pre-Flight Check

Before writing code, verify:
- [ ] Bootstrap done: run the bootstrap command block from HANDOFF-PROMPT.MD
      (init script / dev-server + sanity check). Never burn context figuring
      out setup by trial and error.
- [ ] Required dependencies are installed
- [ ] Database is running and migrations are applied
- [ ] Previous tasks in the chain are complete
- [ ] You understand the task's acceptance criteria
- [ ] You know the task's machine-checked definition-of-done commands
      (from BLUEPRINT task entry / CONSTRAINTS.md `Checked by` / `check:task`)

### Step 3: Implement with Pattern

Write the code following these patterns:

**Controller Pattern:**
```typescript
// Always: validate input → call service → handle errors → return response
export async function handleCreateOrder(req: Request, res: Response) {
  // 1. Validate input
  const body = orderSchema.parse(req.body);
  
  // 2. Call service (never put business logic in controller)
  const order = await orderService.create(body, req.user.id);
  
  // 3. Return structured response
  res.status(201).json({ data: order });
}
```

**Service Pattern:**
```typescript
// Always: check preconditions → execute → handle errors → return result
export async function createOrder(data: CreateOrderInput, userId: string) {
  // 1. Check preconditions
  const user = await prisma.user.findUnique({ where: { id: userId } });
  if (!user) throw new AppError('USER_NOT_FOUND', 404);
  
  // 2. Execute business logic
  const order = await prisma.order.create({
    data: { ...data, userId, status: 'pending' }
  });
  
  // 3. Return result (never return raw prisma result to controller)
  return formatOrder(order);
}
```

**Error Pattern:**
```typescript
// Always: specific error types with HTTP status codes
export class AppError extends Error {
  constructor(
    public code: string,
    public statusCode: number,
    message: string
  ) {
    super(message);
  }
}

// Error codes: USER_NOT_FOUND, ORDER_INVALID, UNAUTHORIZED, etc.
```

### Step 4: Validate Immediately

After writing code, run validation:

```bash
# Type check
npx tsc --noEmit

From susantosanto/config-opencode, MIT.

More from susantosanto/config-opencode