# v0.7.0 — quick test drive

Export a backup, replace files in the same unpacked extension folder, reload at `chrome://extensions`, and confirm Settings shows **v0.7.0**. Do not remove the extension or clear its data.

## Check minimal tracking — no Gemini needed

1. On Today or Settings, inspect **Website time today**. It starts off. Click **Enable website tracking** and approve tab/idle access. Denial must leave tracking off; no Gemini connection is required.
2. Browse one ordinary website for about a minute, keeping the browser focused and interacting normally. Close the panel during part of this test. Reopen it and choose **Refresh totals**; its hostname and elapsed time should appear, not its full URL or title.
3. Switch to another website, then another Windows app. Only the foreground browser site's time should accrue. Refresh after returning; the time away must not be backfilled.
4. Stop input for over a minute, lock the device, or focus a private window. Counting should pause once the relevant idle/focus event is detected. Passive reading/video without input can be undercounted; this is an intentional MVP limitation.
5. Pause tracking, browse, and refresh: saved totals should remain unchanged. Re-enable only if desired. Clearing requires confirmation, removes all tracked days, and pauses tracking without deleting tasks or memory.
6. For a recovery check, suspend the computer for over two minutes or restart Chrome. Resume/refresh: no sleep or closed-browser time should be added. Up to one unfinished minute may be lost.

Snapshots are refreshed manually in this minimal UI; the background counts while the panel is closed. Do not enable debug text or spend a Gemini request merely to test tracking. Report any issue with the window/tab sequence and approximate times; the Gemini debug export excludes browsing data.

## Check the model dropdown

1. Open **Settings → Gemini connection**. If connected, **Gemini model** should already be visible as a dropdown, with the active model labeled separately. If disconnected, choose **Set up Gemini** to see the setup dropdown.
2. Pick a listed model. This must not send a request or switch the active connection yet.
3. Choose **Custom model ID…**. An ID field should appear. Switch back to a listed model: the custom field should hide and not affect submission. In setup, changing the dropdown must not clear a key already pasted into the form.
4. When ready to spend one test request, choose **Use & test model**, or **Connect & test** during setup. A successful test selects that model; a failed model change keeps the current connection.

The catalog and its last-checked date are unchanged in this patch. No automatic model discovery or background connection tests are added.

## Check the review-dialog fix

After your next normal capture with **Allow multiple focus blocks** enabled:

1. Click **Task details and assumptions**. It should expand to show Notes, Deadline, Energy, Priority, and the returned assumptions. If the model supplied none, it should say **No assumptions were provided for this task.** Optional notes/deadlines can still be blank.
2. Collapse it, then focus the summary with Tab and use Enter or Space to expand it again.
3. Toggle **Include block** using both its checkbox and label. The selected-block count should follow your choice; nothing should be saved yet.
4. Edit a task field and use the **Add selected blocks** button (with its displayed count). One confirmation should save the selected blocks; the timer should stay idle.
5. On another draft, check that the close button or a click outside the dialog discards it, but clicking inside does not.

Expanding, editing, selecting, saving, and discarding a draft do not send another Gemini capture request. The automated regressions check click cancellation and handlers; native pointer/keyboard behavior still needs this Chrome check.

## Start without API calls

1. On Today, leave **Planning assistance → Guide me · I make the plan** selected.
2. Expand **Write my task or focus block**. Write a task yourself, choose its minutes, and save. It should appear in Open tasks without a schedule or running timer.
3. Choose **A focus block at my chosen time**, select that task, and enter a future start and your own duration. Save: one manual block should appear in Plan; the timer must remain idle.
4. Try another block at the same time. It should refuse the overlap and retain the form.
5. In Settings, add an explicit memory. Edit it and try Undo under Recent memory changes. Custom Instructions must remain unchanged.

## Try Gemini sparingly

Each action below uses one request. Stop after the feature that interests you most.

1. Type a planning question under **Optional Gemini brainstorming** and choose **Help me think it through**. Questions/ideas should appear; your manual form must remain unchanged.
2. Switch to **Suggest my next step** and ask once. Existing tasks should not be duplicated or renamed. **Review focus setup** must open review, not start a timer.
3. In the connection card on Today or Settings, use the visible **Gemini model** dropdown. Select only the model you want, then **Use & test model**. No key re-entry is needed. Failure should keep the prior model.

## Try asynchronous memory

1. Read the disclosure under **Settings → Saved memory → Learn from my activity**, then enable it if you accept sending that data.
2. Add an observation for learning, for example “A short reading session felt manageable today.” Or save/edit a task or record a timer after opting in.
3. Choose **Refresh memory now** once. The result may be tentative, an observation, a conflict, or no change. One note need not establish a lasting preference.
4. Inspect **Why this is saved** and **Recent memory changes**. Confirm an entry only if you agree; otherwise edit or forget it.
5. Request planning help again. **What informed the last Gemini request?** should identify relevant selected memories.
6. Forget a test entry and check that it and its supporting learning evidence disappear. Forget cannot be undone; original tasks/timers and exported files remain separate.

Automatic learning runs near the configured local workday end, or catches up after reopening. Chrome must be running; it does not wake a sleeping device. Active timers or interactive Gemini work defer automatic updates. A manual update uses that day's automatic slot; all memory calls share a three-per-local-day cap. Failure retains evidence and imposes a 15-minute backoff before a manual retry.

## Report an issue

Open **Settings → Gemini diagnostics → Export debug report** and attach the JSON to the development chat. Text is excluded by default. To share the actual prompt, memories/evidence, and provider error, enable **Include prompt and custom instructions in debug reports** before one retry, then inspect the export. Keys remain redacted.

Mention the mode, action, result, and what you expected. A screenshot helps with layout issues.

Automated tests use simulated DOM/storage/browser APIs and mocked provider responses. Real Chrome permission prompts, tab/focus/idle events, and visual behavior still need this walkthrough. Whole-device and native-app tracking are not present.