Agentic Engineering

# Agentic Engineering

Use this skill for engineering workflows where AI agents perform most implementation work and humans enforce quality and risk controls.

## Operating Principles

1. Define completion criteria before execution.
2. Decompose work into agent-sized units.
3. Route model tiers by task complexity.
4. Measure with evals and regression checks.

## Eval-First Loop

1. Define capability eval and regression eval.
2. Run baseline and capture failure signatures.
3. Execute implementation.
4. Re-run evals and compare deltas.

**Example workflow:**
```
1. Write test that captures desired behavior (eval)
2. Run test → capture baseline failures
3. Implement feature
4. Re-run test → verify improvements
5. Check for regressions in other tests
```

## Task Decomposition

Apply the 15-minute unit rule:
- Each unit should be independently verifiable
- Each unit should have a single dominant risk
- Each unit should expose a clear done condition

**Good decomposition:**
```
Task: Add user authentication
├─ Unit 1: Add password hashing (15 min, security risk)
├─ Unit 2: Create login endpoint (15 min, API contract risk)
├─ Unit 3: Add session management (15 min, state risk)
└─ Unit 4: Protect routes with middleware (15 min, auth logic risk)
```

**Bad decomposition:**
```
Task: Add user authentication (2 hours, multiple risks)
```

## Model Routing

Choose model tier based on task complexity:

- **Haiku**: Classification, boilerplate transforms, narrow edits