Build Loop Test Areas#

This file holds the broad coverage checklist and first human test for build-loop-tests.md.

Required Test Areas#

The Build Test Fix loop must create and run tests for:

• database writes
• MCP tool contracts
• agent logging proof
• agent feedback proof
• stored feedback retrieval
• planned work logging and start gating
• task cancellation with audit trail
• stdio MCP plan, start planned, and cancel lifecycle
• blocker and decision dashboard visibility
• planned work and risks in agent team-state responses
• recent completions in agent team-state responses
• task check before start
• scope check before start
• task-to-workstream linking with ownership checks
• workstream dashboard linked task counts and recent task names
• overlap warning creation
• overlap recall/precision and Performance quality gate
• stale completed work ignored by pre-start overlap checks
• start rejection without confirmation
• start success with confirmation
• WorkOS AuthKit login and local-dev tenant onboarding
• MCP developer token mapping
• cross-developer task mutation rejection
• cross-developer workstream and note mutation rejection
• local-dev identity selection through the dev-login API test path
• MCP token inventory and revocation after refresh
• demo token cleanup
• usage event writes
• session duration tracking foundation
• session heartbeat and stale status
• local repo snapshot capture without file contents
• local repo snapshot conflict warnings for overlapping dirty or unpushed files
• agent-to-agent coordination claims, questions, replies, handoffs, and releases
• idempotent task and workstream completion retries
• completed work cannot be reopened through update tools
• parallel work audits for already-running and recently completed duplicate work
• agent team-state warnings from persisted parallel audit conflicts
• no stale team-state warnings after both sides of an audit conflict are done
• cross-developer session heartbeat/end rejection
• dashboard metrics from real records
• Workstreams planned-work data
• Weekly Summary planned work data
• Coordination risks, decisions, code claims, and peer messages data
• Workstreams data
• Activity completion data
• Activity audit data
• Weekly Summary generation

If the app has a browser UI, the loop must include a browser test for the main user journey.

First Human Test#

When a human opens the app in the morning, they should try this first:

1. Start the server.
2. Open Workstreams.
3. Run the duplicate-work test scenario.
4. Confirm the overlap warning appears.
5. Try starting without confirmation and confirm it fails.
6. Start with confirmation and confirm it appears in the UI.
7. Confirm the MCP response told the agent what to do.
8. Confirm the task and warning were saved.
9. Confirm usage/session events exist.
10. Confirm Performance changed.
11. Run Audit parallel work.
12. Confirm the Parallel Audit panel shows both overlapping records.
13. Refresh and confirm the latest audit still appears.
14. Generate the weekly summary.
15. Confirm structured summary data is visible.
16. Confirm Parallel audits is included.
17. Confirm summary generation does not create a new MCP token.
18. Complete work through MCP and confirm Activity shows the result.
19. Cancel planned work through MCP and confirm the dashboard removes it from Workstreams.

If this works, the customer-readiness slice is real.