NoWorkingOffice AI Studio Control Plane

About

NoWorkingOffice AI Studio Control Plane is a local-only, evidence-driven workflow surface for complex multi-agent studio work. It keeps exact lane state, evidence level, risk, stop reason, next safe step, reversible staging, human authority, and result closeout visible in one place. This Build Week candidate turns NoWorkingOffice's pre-existing operating doctrine into a new runnable Vue product surface using anonymous fictional data. It performs zero external actions and always stops at PUBLICATION_GATE_NOT_ACTIVE. A verified 2:45 silent recording is prepared as the project's unlisted demo recording; manual upload and URL receipt are separate facts and are not claimed in this story.

Inspiration

Multi-agent work often fails between execution and judgment: one agent says a task is done, another leaves an artifact, a third is waiting for approval, and the operator must reconstruct what is actually true. Before Build Week, NoWorkingOffice already had written rules for exact states, local evidence, reversible steps, and human gates. The Build Week goal was to turn those rules into a product surface that someone could understand and operate in under three minutes.

What it does

The demo presents four fictional lanes with current state, goal, risk, stop reason, and next safe step. In the interactive lane, a synthetic brief moves from SOURCE_ASSERTED through a four-rule local verifier, reversible browser-memory packet staging, an explicit human decision, and a structured closeout. The operator can undo staging before the gate. Invalid early decisions are rejected. The final receipt records the artifact, 4/4 verification, human decision, remaining risk, and PUBLICATION_GATE_NOT_ACTIVE.

How we built it

We built an isolated Vue 3 and Vite 6 application around a pure JavaScript state machine. Vue derives the lane rail, execution chain, evidence table, proof ladder, human-gate readiness, and closeout receipt from guarded transitions. The interface uses local system fonts, CSS, and inline SVG; it loads no external asset, analytics, provider SDK, or customer data. OpenAI Codex supported repo-local implementation and verification. Local Browser QA plus Playwright and FFmpeg produced the private demo master; the candidate runtime itself makes no model or provider call.

Challenges

The first challenge was preserving different truths instead of flattening them into one green “done” badge: SOURCE_ASSERTED, ARTIFACT_PRESENT, VERIFIED_LOCALLY, and HUMAN_DECISION_RECORDED had to remain separate in both code and UI. The second was making safety interactive through a real browser-memory stage/undo path before the human gate. The third was keeping the Build Week delta honest through a contribution ledger that credits the pre-existing doctrine and isolates the new runnable product work.

Accomplishments

The complete workflow runs locally in under three minutes. Domain tests reject invalid transitions and verify the stage/undo/decision/closeout path. Desktop 1440×1000 and mobile 390×844 browser QA passed with zero relevant console warnings or errors and no page-level horizontal overflow. We also produced one verified 165-second H.264 1080p silent master with burned-in captions, six chapter keyframes, a contact sheet, and a verified manifest/SHA chain. These are implementation facts, not customer or commercial outcomes.

What we learned

The most useful control-plane feature is not another execution button. It is the ability to answer six questions at every step: What exactly is true? What evidence supports it? What is still blocked? What is the next reversible step? Which action needs a person? What receipt closes the result? When those answers remain visible, complex AI work becomes easier to review and safer to continue.

What's next

Publication and live integration remain separate gates. The verified local recording is prepared for unlisted sharing, while upload, URL receipt, Save & Continue, and Final Submit remain distinct operator actions outside this story. A later phase could connect real agent runtimes only after authentication, data handling, cost, provider, audit, and rollback controls are defined and approved.

Current boundary

This remains a local-only candidate using anonymous fictional data. No public app URL, public repository, deployment, live agent/provider integration, customer data, customer outcome, revenue, ROI, conversion, campaign result, or platform acceptance is claimed. The story does not assert upload status. Final Devpost submission remains inactive.

Built With

• css
• javascript
• openai
• svg
• vite
• vue