One board. Two ways to move work forward.
The story
A small team is preparing a Friday launch. Nobody needs another status meeting or a second task list hidden in a chat. They need to know what happens next, who owns it, and what is in the way.
Relayboard keeps that work in one place: a clear browser board, permission-aware MCP actions for an assistant, and an activity trail that makes changes inspectable. Dodil supplies identity, durable data, deployment and AI. The source is MIT licensed so a remixer can own an isolated copy.
A two-minute walkthrough
Use only synthetic work and your own app login. Dates below should be selected explicitly for the demo day; do not rely on “Friday” being interpreted the same way everywhere.
| Moment | Show | Why it matters | Evidence status |
|---|---|---|---|
| 0:00 — A clear next step | Create Friday launch. Add Prepare the walkthrough, a short description, yourself as assignee, a due date, and the launch label. | Turn an intention into owned, concrete work. | Real deployed browser/DataK3 round-trip verified. |
| 0:25 — Make progress visible | Move the card from To do to Doing. Use either a pointer drag or the Move selector, then reload. | The interaction is simple; the result is durable and keyboard accessible. | Real desktop drag, keyboard move and persistence verified. Mobile is not yet verified. |
| 0:45 — One source of truth | With an authorized external assistant connected, ask it to find this board and card, resolve the visible card reference, read the current revision, and move the unique match to Done. Reload the browser. | Chat takes a real action on the same record instead of inventing another task list. | Connected ChatGPT owner reads, reference lookup and a revision-bound To do → Doing → To do move/readback/restore passed. This proposed Done scene is not a recorded walkthrough; Claude, broader roles and OAuth internals remain unverified. See scoped evidence. |
| 1:05 — Trust the change | Open Activity. Point out the actor and timestamp, then expand the “View changed fields” disclosure to see the before/after list values. | A team can inspect what changed rather than take an assistant’s word for it. | Real audit rows and transaction conflicts verified. |
| 1:20 — Find the next useful action | Return the card to Doing. Open My week, request a checklist suggestion, and run Briefing. | Focus on work that needs attention, with AI suggestions the person reviews. | Real Kimi K2.6 checklist/summary and Jina Embeddings v4 matching verified; quotas apply. |
| 1:45 — Make it yours | Show the MIT repository, .dodil/remix.yaml, CI checks, and the documented first-sign-in checks. | A builder gets inspectable code and a path to its own bucket, user pool and deployment. | Official validation/ready plan passed. Fresh-organization remix execution is still pending. |
External assistant prompts, after connection
Use the app’s Connect your assistant panel for the actual MCP URL. Complete the assistant’s OAuth sign-in as yourself, and have the pool administrator approve only the verified pending client. Do not paste service-account credentials into an assistant.
- “Show my boards and find Friday launch. Show its cards before changing anything.”
- “Move Prepare the walkthrough to Done, using the revision you just read.”
- “Show the board activity so I can check that change.”
- “What unfinished work is assigned to me over the next seven days? Include overdue and undated work.”
- “What is blocking this card? Suggest a checklist; do not change its description without asking me.”
A refusal or conflict is a real result: refresh the card and ask for confirmation rather than claiming an action succeeded. If the intended client is not connected and verified in your session, omit the assistant scene or label an HTTP/MCP test as a protocol test—not a Claude/ChatGPT demo. The connected ChatGPT test is not proof that another client or remixed deployment is ready.
Reviewed checklist scene — verified on the hosted demo
The hosted release has passed real browser/DataK3 checks for AI review/select/add-to-draft, explicit save, completion with stable step IDs, reload persistence, canceled drafts and rejected stale edits. This is live test evidence, not a recorded walkthrough or proof of external assistant OAuth; see verification details.
On an authorized test board, open the launch card, request a checklist, deselect an irrelevant step, and choose Add selected to card draft. Show that nothing is saved yet, then choose Save card. Reopen, mark one step complete and save again. Reload to demonstrate persistence, the board’s completion count, and the checklist before/after audit. Repeat a draft addition and Cancel to demonstrate that AI text is not silently saved. Use only real resulting screens, not a mockup labelled as proof. The hosted demo has received migration 002; other existing copies need their own approved migration before running this release. Fresh remixes use the complete schema without migrations.
Optional trust appendix
Use separately approved test identities, not real teammates or production projects. An owner may grant member/viewer access to a synthetic board. Demonstrate that a member can change cards, a viewer can read but cannot write, and an outsider cannot read the board. Show that a denied write creates no audit event. Remove the temporary grants and identities afterward; deleting a user is not a promise of immediate access-token revocation.
Our live test exercised these boundaries using three temporary appid users and real HTTP/MCP requests. The accounts and extra records were cleaned up. See verification details and limits, including the remaining OAuth-internals, broader client/role and fresh-org gates. These historical role tests are not member/viewer/outsider ChatGPT-session tests.
Before recording or submitting
- Recheck the final deployed commit and its CI; do not splice screenshots from a different revision into a “live” proof.
- Keep personal email addresses, tokens, settings screens and unrelated organization data out of recordings.
- Verify the remixer’s free quotas before AI calls. Never imply unlimited or cost-free capacity.
- Show actual outputs; do not invent a client connection, successful remix, invitation email, mobile pass or contest award.
- Check current eligibility and deadline. Submit manually through a permitted channel; no automated Freelancer submission is part of this project.