Lumos — Review Before Action

Watch the complete English narrated demo (1:45)

Inspiration

An assistant can misunderstand a request, especially its timing or priority. People need a chance to inspect an action before it changes their task list. Lumos explores a service-visit reminder while keeping the final decision with the person.

What it does

Lumos — Review Before Action is an explicitly labeled Alexa+ web simulation. A user types a supported Turkish task request, reviews the proposed title, priority and time note in the Lumos panel, and approves or cancels. The simulation provides a browser-local event log and JSON export.

This is not a live Alexa+ integration. The demonstrated flow uses no Alexa device, Amazon account, speech recognition, or runtime language model. It uses a narrow deterministic parser, not general language understanding. A separate Ring Playground API verification is documented below; it is not connected to this Alexa+ web flow.

How we built it

An Astro page embeds the existing Lumos panel and calls its proposed-task confirmation bridge. JavaScript parses the supported request and records simulation events in localStorage. Playwright provides the focused end-to-end scenario.

Lumos and its task panel existed before the hackathon. The competition work adds the Alexa+ simulation page, task-proposal parsing, event export, a focused end-to-end test and simulation-specific documentation. The public demo revision is 53fd83aff48806ba07b8cf0bd9064d76cc3f081d; the jury runbook is pinned in d7801ab1fc1be8e71b81f84a73a39d05590abdbf.

Challenges and verification

The demo needed reliable readiness signaling between its page and embedded panel, and matching confirmation-dialog selectors in its browser test. These were corrected in the pinned demo revision. The development test report records ALEXA_PLUS_SIM_E2E_RESULT: PASS at that revision. This is local web-simulation evidence, not a live Amazon integration test.

The public video demonstrates the review, cancel, approve and event-export flow. It is approximately 1 minute 45 seconds, with synthesized English narration and English subtitles accompanying the Turkish interface. It includes inspection of the actual exported JSON event log. The screen recording is from September 16, 2026; this edition adds narration without changing the application code. A full English translation of the jury runbook is included in the submission’s Testing instructions.

Try it

Source branch: https://github.com/candasoz01-cmd/lumos-core/tree/amazon-alexa-plus-demo-ready-20260916

Jury instructions: https://github.com/candasoz01-cmd/lumos-core/blob/d7801ab1fc1be8e71b81f84a73a39d05590abdbf/docs/analysis/amazon-alexa-plus-simulation-2026.md

Use Node.js 22.12 or later: git clone https://github.com/candasoz01-cmd/lumos-core.git cd lumos-core git checkout 53fd83aff48806ba07b8cf0bd9064d76cc3f081d cd ui npm install npx playwright install chromium npm run build npm run e2e:alexa-plus-sim npm run preview

Open http://127.0.0.1:4321/alexa-plus-simulasyon and use the prefilled Turkish sample. The automated check's expected result is ALEXA_PLUS_SIM_E2E_RESULT: PASS.

Limitations and next steps

The parser supports a narrow set of Turkish task requests. Event records include extracted task fields and can contain personal text; browser-local storage is not anonymous storage. We have not measured usability or customer adoption. A real Alexa+ integration would require separate access, implementation and verification.

Additional Open Source contribution

Alongside the Alexa+ simulation, we are entering the Open Source mini challenge with a separate contribution: https://github.com/candasoz01-cmd/lumos-core/pull/839. Created and merged on September 7, 2026, it improves hosted-chat handling of transient OpenAI errors while preserving the existing total time budget and quota-error fallback. It adds three regression cases. All 28 hosted-chat tests passed again on September 19 at contribution head f485fdf5b917b8a592aa914e124018fb5b2a867c, using stubbed provider responses. This contribution is independent of the simulation, which still makes no runtime language-model calls. Source and tests are public under Apache-2.0.

Development assistance and recent verification

Codex was used to inspect the code and requirements, fix the demo readiness and test selectors, prepare submission material, and rerun verification. AI-assisted development is distinct from the deterministic simulation's runtime. On September 19, the public demo branch was verified at d7801ab1fc1be8e71b81f84a73a39d05590abdbf (documentation over demo revision 53fd83a); its build and focused Alexa+ browser test passed again. The build emitted a pre-existing CSS minifier warning about color:{ACCENT}; this was not treated as a clean warning-free build.

Ring Partner API verification — September 19, 2026

A separate read-only Python integration now calls the official Ring Partner API through the Lumos integration registry. GET /v1/devices returned HTTP 200 with one official Developer Playground test device; GET /v1/devices/{id}/status returned HTTP 200 with online=true. The remote API requests are real. The device is simulated, and this is not physical Ring hardware or production authorization evidence.

The 56-second English narrated, captioned video records the actual operator verification view. It is not customer UI. No token is shown, and no camera recording, device control or automatic action is demonstrated. Ring data does not currently drive the Alexa+ web task-approval flow.

Ring video: https://youtu.be/VpC01p-KWfg

Ring source: https://github.com/candasoz01-cmd/lumos-core/tree/codex/ring-playground-live-20260919

Reproduction instructions: https://github.com/candasoz01-cmd/lumos-core/blob/codex/ring-playground-live-20260919/docs/analysis/ring-playground-live-20260919.md

The Ring feature includes the JSON:API-shaped offline sandbox, read-only live adapter, registry integration and tests. The adapter fails closed without a token. At publication preparation, 26 focused tests passed; the full pre-commit suite passed 2246 tests with 3 skips. These automated tests use injected/stubbed responses; live API evidence is the separate recorded run. Repeat live verification requires a fresh Playground token. English narration is synthetic. Codex helped verify the existing adapter, capture the real requests and prepare the evidence.