Roadmap
Shared direction
A roadmap built around proof
Ten workstreams, 57 proposed tasks, and one goal: accepted delivery that survives real failures.
Start with the next eligible proof
First verify the installed next-action contract and the existing continuation/reply owners. Then qualify the corrected implementation flow. Execute the eligible DZADAP pilot only when its exact dependencies and host authority are current. Current disk and quota reserves are prerequisites, not historical assumptions.
Wave 0
Establish current truth
Storage capacity, installed Worker routes, adapter receipts, assistant consumers and token baselines. Reuse current pilot preparation.
Wave 1
Close communication and continuation
Qualify the landed next-action contract, same-task continuation, A2A completion and owner-aware lifecycle routes. Reuse memory protections.
Wave 2
Prove execution
Accept the corrected native implementation flow and a separately authorized eligible DZADAP pilot. Qualify assistant/package dependencies.
Wave 3
Accept a real requirement
Deliver the selected real slice, consumer journeys, shared coordination and lifecycle rows. Settle actual cost.
Wave 4
Scale from measured acceptance
Only then expand programmes, principals and consumers with cumulative limits, reserves and evidence.
Take the plans with you
Download all ten proposed plans (ZIP)
Ten Markdown workstream plans plus a master README. The package preserves 57 task IDs, acceptance criteria, dependencies and deferred scope. These are proposals and qualification follow-ups, not execution admissions.
Workstream plans
| Area | Proposed priority | Tasks |
|---|---|---|
| DZADAP | P0 | 6 |
| AI Assistant | P1 | 6 |
| A2A | P0 | 5 |
| Worker Coordination | P0 | 6 |
| Token Optimization | P0 | 6 |
| Delivery Control Flow | P0 | 6 |
| Steward Improvements | P0 | 5 |
| Agent Adapters | P1 | 5 |
| Memory / Context / Prompt | P1 | 6 |
| Disk Space / Git Worktrees | P0 | 6 |
Shared dependencies
Storage and installed-state checks support every live proof. Steward, Worker and internal A2A form the continuation loop. Qualified adapters and scoped context support the assistant and delivery runtime. Delivery acceptance enables the DZADAP vertical slice; measurement then determines whether scaling is justified.
A blog task list does not reserve repository ownership, change a task status, authorize spend or replace a controlling admission. Existing owners keep their work.