feat: design system release

.github/workflows/content-publish-automerge.yml

@@ -131,7 +131,7 @@ jobs:
 
                   # Key off the `ci-success` check run, NOT `workflow_run.conclusion`.
                   # `ci-success` is the single check the "Protect Prod" ruleset gates
-                  # on; the run conclusion also folds in advisory jobs (e2e). Gate on
+                  # on; the run conclusion also folds in advisory jobs (ds-shots). Gate on
                   # exactly what the ruleset gates on.
                   #
                   # `Tests` runs on `pull_request` (and `push` for dev/main), so a

.github/workflows/pr-assets-cleanup.yml

@@ -0,0 +1,31 @@
+name: PR assets cleanup
+
+# The ds-shots comment in tests.yml renders its before/after PNGs from a
+# `pr-assets-<number>` branch, because GitHub markdown cannot show a workflow
+# artifact. Delete that branch when the PR ends, or the branches pile up — 20+
+# hand-made ones were still on the remote in August 2026.
+
+on:
+    pull_request:
+        types: [closed]
+
+permissions:
+    contents: write
+
+jobs:
+    delete-assets-branch:
+        name: delete-assets-branch
+        runs-on: ubuntu-latest
+        steps:
+            - name: Delete pr-assets-<number>
+              env:
+                  GH_TOKEN: ${{ github.token }}
+                  REPO: ${{ github.repository }}
+                  PR: ${{ github.event.pull_request.number }}
+              run: |
+                  # Most PRs never make this branch — a docs PR skips ds-shots,
+                  # and a fork PR cannot push. A missing branch is normal, so a
+                  # 422 from the delete is not a failure.
+                  gh api -X DELETE "repos/$REPO/git/refs/heads/pr-assets-$PR" \
+                      && echo "deleted pr-assets-$PR" \
+                      || echo "no pr-assets-$PR branch to delete"

.gitignore

@@ -60,6 +60,8 @@ next-env.d.ts
 
 # local files
 playground/
+# ...except the design-system doc-site Playground route, which is a real tracked route
+!src/app/(mobile-ui)/dev/ds/playground/
 
 # yarn and npm lockfiles
 package-lock.json
@@ -93,6 +95,7 @@ e2e/__results__/
 e2e/__baseline__/
 e2e/__report__/
 e2e/__snapshots__/
+e2e/__shots__/
 e2e/.auth/
 
 # native app signing

.prettierignore

@@ -17,5 +17,9 @@ src/components/TransactionDetails/__tests__/fixtures/render-baseline.json
 src/types/api.openapi.json
 src/types/api.generated.ts
 
+# generated from globals.css @theme via `pnpm gen:ds-tokens` — stable
+# JSON.stringify formatting so a prettier bump can't flip the drift gate.
+src/app/(mobile-ui)/dev/ds/foundations/tokens.generated.ts
+
 # static single-file quiz (hand-tuned inline CSS/JS; not prettier-formatted)
 public/onboarding-quiz/

docs/TESTING.md

@@ -6,8 +6,14 @@ Full testing strategy (pyramid, CI/CD, post-release monitoring) lives in the mon
 ## Quick commands (this repo)
 
 ```bash
-npm test                              # Jest unit + component (710+ tests, ~25s)
-npx playwright test --project=mobile  # E2E smoke (49 tests, ~8 min)
-npx tsx e2e/scripts/generate-report.ts --save-baseline  # Visual regression baseline
-npx tsx e2e/scripts/generate-report.ts                  # Compare against baseline
+npm test                                        # Jest unit + component (~30s)
+NEXT_PUBLIC_VERCEL_ENV=preview npm run build    # both Playwright configs need this build
+npm run test:e2e:regression                     # behaviour specs in e2e/flows (~1 min)
+npm run test:visual:capture                     # one PNG per fixture per width
+npm run test:visual:diff <before> <after>       # compare two capture directories
 ```
+
+Browse the fixtures at `/dev/fixtures`, or open one with `<route>?__fixture=<name>`.
+
+Anything that needs a real backend, provider or chain goes to the Nutcracker
+harness in mono (`engineering/qa`), not to Playwright here.