ai-driven-dev Backlog
flowchart LR
intent([ask, create, manage, review, or refine]) --> inspect
inspect -->|"ask"| answer([answer from the read model])
inspect -->|"create"| triage --> route
inspect -->|"manage"| route
inspect -->|"refine"| assess
inspect -->|"review"| review
route --> decide
assess --> decide
review --> decide
decide -->|"revise"| route
decide -->|"no change"| done
decide -->|"authorize"| apply --> verify
verify -->|"invalid"| route
verify -->|"coherent"| done([backlog])
Actions
Run the flow above. Read only the next action file.
| Action | Does |
|---|---|
| inspect | resolve the scope, event, and authority |
| triage | classify intake and detect existing work |
| review | find actionable backlog health improvements |
| route | delegate proposed artifact changes |
| assess | challenge refinement from three viewpoints |
| decide | reconcile proposals and confirm authority |
| apply | delegate authorized writes |
| verify | prove the resulting graph is coherent |
Transversal rules
- Keep product, lifecycle, and backlog decisions with the user.
- Delegate artifact work to its owning capability, resolved by purpose and verified as installed before delegating.
- Store each relation once, in its owning artifact.
- Ask natural questions; never expose internal routes, checks, or unchanged state.
- Change only authorized artifacts and fields.
Say when this orchestration is over
Once the flow has reached done — an answer given, no change needed, or verify reporting a coherent graph, and only then, run:
echo "aidd:step-end aidd-orchestrator:02-backlog"
No host reports when an orchestration finished. A skill call's own result comes back in a tenth of a second, which is the dispatch and not the completion, so a measurement that never hears this ends the run at the next pause — and an orchestration is mostly pauses. Everything it drove afterwards then reads as work that belonged to nothing.