The Stage Director's Flow
The Director's Flow is an optional attention board for a solo director working with an engineering agent. Use it when several outcomes compete for attention or work spans several slices. One bounded change can stay in conversation and use the ordinary delivery loop.
A Small Attention Board
| View | What belongs here |
|---|---|
| Ideas | Raw Cues, problems, observations, and obligations worth remembering |
| Approved | Outcomes the agent may implement without inventing product intent |
| Active | Work currently consuming engineering or integration capacity |
| Needs a human | A concrete product decision, fair play surface, policy decision, or release proposal |
These views are suggestions, not required states or transitions. Use existing paper notes, Markdown, or a tracker. Completed work can leave the board with a commit or useful finding; do not maintain duplicate status in several places.
A Cue is a raw idea. Ready means approved for implementation, not fully specified. A new or materially changed gameplay premise needs a human-approved hypothesis before becoming Ready. A clear ordinary fix may already be approved by its request. Do not add a second approval step or reopen settled design.
The Work Item
When persistent context helps, use the five-field starter:
| Field | Useful content |
|---|---|
| Outcome | The change or answer wanted and why it matters |
| Scope | Authorized boundary, preserved behavior, protected state, and non-goals |
| Checks | Production consumer, mechanical evidence, and human inspection needed |
| Decision | Accepted direction, unresolved choice, rejection, or play feedback |
| Checkpoint / next step | Commit or current result, remaining limitation, and next action |
Only work with real coordination needs adds path ownership, integration order, worktree coordinates, or a handoff inside these fields. Do not refine every idea, estimate fictional deadlines, or require a release impact for every edit.
Attention And Authority
The director chooses product intent, priorities, acceptable tradeoffs, subjective acceptance, and release. The agent completes authorized engineering, verifies material claims, and makes coherent local commits after relevant mechanical checks pass. A commit can precede human play or visual acceptance; it records recoverable engineering state, not the director's verdict.
Keep active work within actual review and integration capacity. For a new player-facing premise, complete one playable loop and obtain human feedback before broad expansion. Independent maintenance may proceed when it is useful and does not interfere with protected state or the pending decision. See gameplay guidance for the play boundary.
When a decision is needed, bring a concrete result: a playable slice, alternatives with consequences, preserved failure evidence, or a finished release proposal. Ask only for the unresolved decision. After feedback, revise, accept, reject, or defer the affected work. Preserve accepted fragments and explicitly rejected routes. A technically successful experiment can still deserve rejection.
Useful Measures
Look at human active time, time to a fair inspection or play surface, useful acceptance or rejection, rework, escaped defects, and method-maintenance effort. Use brief observations in the existing evaluation context. Board throughput, agent activity, and commit counts are not product value.
The former seven-state board is retained in historical decisions. Existing boards can keep their labels; updating STAGE does not require migrating project artifacts.