Relationship-Agent UAT Checklist

Use this checklist to validate that CoWork OS behaves as a personal relationship agent, not only a task executor.

Preconditions

1. App is running with a configured LLM provider.
2. Workspace is selected.
3. Build health already passes:
   • npm run type-check
   • npm run build

Test 1: Strategy Prompt (Answer First)

Input

I built you, how should I position you, who should I target, and what should I aim to achieve?

Expected

1. A direct answer appears early.
2. The agent may continue deeper execution, but should not withhold the main answer.
3. Task final state is completed.

Test 2: Mixed Mode (Talk + Act)

Input

Give me a quick positioning recommendation first, then produce a deeper GTM strategy document.

Expected

1. Initial concise recommendation.
2. Follow-on execution for deeper artifact.
3. Final user-facing completion response.

Test 3: Timeout Recovery (No Silent End)

Setup

Use a prompt likely to create a long synthesis step.

Expected

1. If timeout occurs, logs show recovery behavior.
2. Task still returns a best-effort final response.
3. Task should not terminate without user-facing output.

Diagnostic Logs

• Good signal:
  • Task cancelled - not logging as error (reason: timeout) with recovery/finalization events.
• Bad signal:
  • cancellation log without any final user response.

Test 4: Relationship Memory Capture

Input

Call me almarion. I prefer concise responses. Remind me to send investor update tomorrow.

Expected

1. Relationship memory includes identity and preference items.
2. Commitment item exists and is open.
3. Due-soon reminder is returned within the configured window.

Test 5: Relationship Memory Controls

Actions (Settings > Memory System)

1. Edit a relationship item.
2. Mark a commitment done.
3. Reopen the same commitment.
4. Forget one item.

Expected

1. UI updates immediately.
2. Changes persist across restart.
3. Due-soon list reflects status changes.

Test 6: Image Attachment in Task Creation

Input

Create a new task with a JPEG or PNG image attached and the prompt: Describe what you see in the attached image.

Expected

1. The LLM response references specific visual content from the image (not generic text).
2. No "Image skipped" warnings appear in the activity log.
3. Follow-up messages with additional image attachments are also processed correctly.

Test 7: Shared Context Safety

Setup

Run equivalent prompts from private and shared channel contexts (if configured).

Expected

1. Private context can use relationship memory normally.
2. Shared context follows memory isolation policy unless explicitly trusted (allowSharedContextMemory).

Pass Criteria

Release is accepted when all seven tests above pass with no silent task termination and no regression in completion behavior.