Spec-driven development
All Aucert engineering work follows a spec-driven workflow. The formality scales with task size.
Task sizes
| Size | Time | Planning approach |
|---|---|---|
| Small | < 1h | Plan in conversation (Plan mode), then execute |
| Medium | 1-4h | Write .tasks/backlog/ file → agent generates spec in .tasks/drafts/ |
| Large | Multi-day | Detailed backlog file → comprehensive spec → both founders review |
Task file locations
| Directory | Purpose |
|---|---|
.tasks/backlog/ | Raw intent files awaiting spec generation |
.tasks/drafts/ | Generated specs awaiting review |
.tasks/active/ | Approved specs ready for execution |
.tasks/archive/ | Completed tasks |
Context loading
Task files specify which context to load:
context_load:
l1: [ARCHITECTURE.md, CONVENTIONS.md]
l2: [backend/.context/BACKEND.md]
l3: [backend/.../domain/generation/spec/GENERATION.md]
What's next
- Drift capture — How we track implementation drift from specs
- 5-layer deep dive — Pipeline architecture