Check Ci

Continuously fetch, fix, and push until CI is fully green on the current branch.

Uses `~/ci.sh`, which fetches the latest PR CI run for the current branch, polls
until it completes (or a failure is detected early), downloads per-job logs, and
classifies each failure as locally verifiable or not.

## Outer loop

Repeat until done (max 5 push cycles before stopping):

1. Run `bash ~/ci.sh` **in the background** (`run_in_background: true`) — it blocks
   and polls for up to 30 minutes. You will be notified when it finishes; read its
   output to get the `STATUS:` line and failure details.
2. Check the `STATUS:` line:
   - `passing` / `no_runs` → stop, CI is green
   - `timeout` → stop with error: "CI timed out after 30 minutes"
   - `failed` → fix (see Fix steps below), then return to step 1
3. After 5 push cycles with no green result, stop and report what's still failing.

## Script output format (when STATUS is `failed`)

STATUS: failed RUN_ID: OUT_DIR: /workspaces/neon/.tmp/neon-ci//

=== FAILED JOBS ===

JOB: web-ci / Server Unit Tests (3) LOCAL: yes LOG: /workspaces/neon/.tmp/neon-ci//server-unit-tests-3-.txt

JOB: shared / check-translations LOCAL: yes LOG: /workspaces/neon/.tmp/neon-ci//check-translations.txt

JOB: merge-gatekeeper LOCAL: no LOG: /workspaces/neon/.tmp/neon-ci//merge-gatekeeper.txt

`LOCAL: yes` means the failure can be reproduced and verified locally.
`LOCAL: no` means you can read the log and attempt a fix, but cannot verify it runs green before pushing.

## Why certain jobs are LOCAL: no

- **Staging E2E** — runs against a deployed staging environment; no local equivalent.
- **Console Cypress** — needs a separately seeded dataset and the console dev server
  on port 3003; not straightforward to set up locally.
- **Docker** — builds and pushes images; requires a Docker daemon and registry credentials.
- **Gatekeeper** — a GitHub-side merge guard, not a runnable test.

## Fix steps (when STATUS is `failed`)

1. Read each job's `LOG:` file to understand the error.
2. Fix all failing jobs in one pass before re-verifying anything.
3. For `LOCAL: yes` jobs, verify fixes by running the local command and redirecting
   output to a temp file — never rely on scrollback:
   ```bash
   yarn workspace server test > .tmp/test-out.txt 2>&1 || true
   # then Read or grep .tmp/test-out.txt for failures

4. For LOCAL: no jobs, fix based on log content only; skip local verification.
5. After all LOCAL: yes commands pass, run lint/typecheck for every affected workspace — fixes can introduce type or lint regressions.
6. Stage only changed source files, commit, and git push.
7. Return to the outer loop.

Local command by job type

Job name pattern	Local command
Unit Tests (storefront)	yarn workspace storefront test
Unit Tests (console)	yarn workspace console test
Unit Tests (neon-dash)	yarn workspace neon-dash test
Unit Tests (neon-js)	yarn workspace neon-js test
Unit Tests (checkout)	yarn workspace checkout test
Server Unit Tests	yarn workspace server test
check-translations	yarn workspace <app> run check-translations
Cypress integration tests (checkout, ...)	see below
Cypress integration tests (neon-js, ...)	see below

If a job doesn't match, derive the workspace from the job title.

Running Cypress integration tests locally

Prerequisite: The test checkout session must exist in the local DB. If the backend returns 404 for the fixture URL, run the seed first:

yarn run db:setup

This is idempotent (upserts) and safe to re-run against the existing local DB.

Both the API server (port 8080) and the relevant app dev server must be running. Checkout and neon-js dev servers are already up in this environment (ports 3001 and 3010).

Read the failing log to identify which spec file failed, then run it directly — omit --record since Cypress Cloud credentials are not available locally:

# checkout
cd apps/checkout && npx cypress run --spec 'cypress/integration/<failing-spec>.cy.ts'

# neon-js
cd apps/neon-js && npx cypress run --spec 'cypress/integration/<failing-spec>.cy.ts'

Fix rules

• If the same root cause appears across multiple jobs, fix it once — don't duplicate.
• Never commit files that look like secrets or generated artifacts.