Work lifecycle

This is the full collaboration loop for one batch, from a human intention to a merged change. The agent runs the mdBoard actions; the humans make the decisions.

At a glance

spark → refine → pitch → accepted → roadmap → breakdown → doing → handover → completed
                                                                           │
                                                                       human merge
                                                                           ▼
                                                                    upstream → nightly

Humans decide at the marked gates. The agent works inside the batch.

1. Spark

You: state an intention.

Human: Spark an idea for CSV export. Keep it small: current filters,
       streamed response, owner-only.

Agent: Created #0007 as a spark. No scope invented beyond your sentence.

mdBoard: creates an idea card in spark.

2. Refine

You: shape the intent with the agent.

Human: Ask me the open questions, then write acceptance criteria.

Agent: Three questions: hidden columns, streaming boundary, access.

mdBoard: the card is in refine; the agent may edit content only here.

3. Pitch

You: ask for the consortium decision.

Human: Pitch it at the next consortium review.

mdBoard: the card is in pitch; content is frozen.

4. Consortium decision

The consortium: accepts, rejects or requests changes.

Human: The consortium accepts #0007. Promote it.

Agent: Recorded with you as authoriser. #0007 is accepted and is now
          a batch. Shall I estimate it?

mdBoard: accept-idea changes the kind from idea to batch, state accepted.

5. Estimate and roadmap

You: commit the batch to the roadmap.

Human: Estimate it and put it on the roadmap.

Agent: 12,667 tokens, 1.0 agent hour, about €0.06. Roadmap 3 of 5.
          Confirm?

Human: Yes.

mdBoard: the batch gets an estimate, then moves accepted → roadmap. The roadmap total updates.

6. Breakdown and clarification

You: ask for the decomposition.

Human: Break it down and ask me everything before starting.

Agent: Three tasks created. One clarification meeting is pending:
          two questions about the streaming boundary.

mdBoard: batch roadmap → breakdown; tasks created in pending; a clarification gate is pending.

Answer the questions. The agent records the answers and accepts the gate. The batch does not start until you say so.

7. Start and execute

You: authorise implementation.

Human: Start #0007.

Agent: #0007 is doing. #0007-01 is working. I will commit each task
          and keep one task working at a time.

mdBoard: batch breakdown → doing; the agent works pending → working → done for each task.

The agent is autonomous inside this batch. It does not ask for permission per line.

8. Handover

Agent: #0007 is in handover. All tasks done. Report draft ready.
          I opened the handover review.

mdBoard: batch doing → handover; a review round begins.

9. Review and changes

Reviewers: read the batch, not every task.

Human: Changes required: the stream closes before the final row.

Agent: Recorded the review, returned #0007 to doing, and created
          rework task #0007-04 from finding F1.

mdBoard: handover → doing; rework task in pending; next handover is round 2.

When the rework is done:

Human: I accept the batch. Complete it.

Agent: Owner accept recorded. Report and changelog written.
          #0007 is completed. Next gate: the upstream merge.

mdBoard: handover → completed; report and changelog written.

10. Merge and release

You: prepare, then execute the merge.

Human: Prepare the merge for this iteration.

Agent: The worktree is clean, the merge check passes, no active batch.
          Only you can merge.

mdBoard: the owner merges upstream → nightly. On the release train, the owner merges nightly → stable and tags the release.

Where the meetings fit

The agent keeps pending gates visible in every board brief.

What the human does

• states intent and constraints;
• accepts or rejects ideas;
• selects and estimates roadmap batches;
• answers clarification questions;
• starts implementation;
• reviews and accepts the batch;
• merges and releases.

What the agent does

• refines, estimates, breaks down;
• creates tasks and asks questions;
• implements tasks and commits;
• writes the report and changelog;
• transcribes reviews with provenance;
• reminds you about meetings;
• prepares merge checks and release notes.

What mdBoard does

• enforces the state transitions and authority;
• records events and human authorisations;
• tracks WIP limits and meeting gates;
• generates the board and roadmap;
• validates that completion has evidence.

Next

• Review and acceptance
• Meetings
• Branches and releases
• Estimation and cost