# Build Loop Done Criteria

Agents may claim the overnight loop is done only when this file is satisfied.

## Done Means

The loop is done only when all of these are true:

- the app starts locally
- the UI opens
- MCP tools respond
- authenticated UI login works, or auth gap is clearly documented
- local dev identity can be selected in the UI
- MCP calls map to a developer identity
- MCP task mutation is scoped to the owning developer
- MCP workstream and referenced-note mutation is scoped to the owning developer
- active MCP tokens can be listed and revoked after refresh
- local companion setup installs MCP and Claude Code hook config without storing raw tokens
- installable companion package smoke passes in a clean external repo
- local companion status distinguishes hard setup failures from server reachability warnings
- local companion snapshot records dirty and unpushed repo state without file contents
- local companion watch can repeat snapshots, telemetry inbox flushes, and optional session heartbeats
- local companion snapshot warns agents about overlapping dirty or unpushed files from other sessions
- duplicate-work warning works
- duplicate warning is sent back to the agent
- stored warnings and blocks can be retrieved by the agent
- overlap start is blocked without confirmation
- block response tells the agent what to do next
- overlap start succeeds with confirmation
- demo scenario does not leave demo MCP tokens in inventory
- structured events are stored
- usage events are stored
- session/task timing fields exist
- planned work is visible before start
- planned tasks cannot bypass check-before-start
- abandoned work can be cancelled with a reason and activity event
- blockers and decisions are visible without generating a summary
- long-running sessions support heartbeat and stale status
- session heartbeat/end is scoped to the owning developer
- metrics show sessions, usage, warnings, and completed work
- Workstreams shows active and planned work
- Activity shows completed work and result summaries
- Activity shows events
- Weekly Summary produces a draft
- Weekly Summary exposes structured payload data
- UI summary generation does not create MCP tokens
- the accuracy gate for overlap recall/precision and Performance read-models passes
- tests pass, or failures are clearly listed
- security checks pass, or failures are clearly listed with severity
- hosted deployment artifact, hosted canary, and strict handoff command exist
- morning report exists

## Not Done Means

The loop is not done if:

- only the UI exists
- only the MCP exists
- data is stored only in Markdown
- overlap warnings are not enforced by the server
- warnings only appear in the UI and are not returned to the agent
- MCP calls do not save usage events
- MCP calls are not tied to a developer identity
- auth is ignored without a clear reason
- the weekly summary is missing
- tests were not run
- `npm run check:accuracy` was not run or failed
- no browser/user-flow test was attempted
- security checks were not run
- the morning report is missing
- any source file is over 300 lines
- documentation is stale
- deferred feature code is present

## Morning Report

At the end of the loop, agents must write `reports/morning-report.md`.

The report should say:

- what was built
- what works
- what tests passed
- what tests failed
- what security checks passed or failed
- what was not finished
- what decisions were made
- what files changed
- what a human should try first
- what auth provider was chosen and why
- whether agent logging was verified
- whether agent feedback/blocking was verified
- whether the accuracy gate passed

## Manual QA Instructions

The loop must leave instructions for a human to try the app.

The instructions must include:

- how to install dependencies
- how to start the server
- how to run `npm run setup` to initialize PostgreSQL and run the local doctor
- where to open the UI
- how to connect an MCP client
- how to use the hosted companion install command, optionally override it, and run `npm exec -- covibe-local setup --base-url <origin>`, manual `snapshot --base-url <origin>`, and `watch --base-url <origin>`
- how to create or configure a developer MCP token
- how to run the duplicate-work scenario
- how to generate the weekly summary
- how to run `npm run readiness:handoff -- https://...` after deploy