Two entries: - **MCP gateway + Persona-runner — end-to-end live smoke** (🟠) Covers M1+M1.5+M2+M3 commits. Unit tests verified ~2600 LOC at the type/shape level, but nothing has ever talked to a real Postgres + mana-auth + Anthropic. 11-step recipe walks through seed → tick → verify in psql, including the encryption-on-wire check (enc:1: prefix in sync_changes, plaintext in web app). - **Persona visual regression — capture first baselines** (🟡) Depends on the smoke run above succeeding (empty personas produce meaningless baselines). Eyeball-check step is explicit — the first PNG IS the reference, no CI can catch "baseline was wrong". Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
10 KiB
Manual / Smoke Test Backlog
Single source of truth for "features that are code-complete + unit-tested but still need a human click-through before release." Distinct from:
test-health.md— automated test suite healthdocs/TESTING_DEPLOYMENT_CHECKLIST.md— CI/CD system pre-deploydocs/plans/shared-space-smoketest.md— detailed shared-space walkthrough (linked below)
An entry lives here as long as the feature hasn't been driven through a real browser / device / account end-to-end. When the smoke test has run green once, delete the entry.
Format
Each entry carries:
- Area — feature surface
- Why it's here — what unit tests do not cover
- Steps — inline if short; a link to a dedicated walkthrough if long
- Shipped? — commit(s) that shipped the code the test would exercise
- Priority —
🔴 release blocker/🟠 important/🟡 nice to have
Open
Data Export v2 — end-to-end roundtrip in a real browser
- Priority: 🟠 important
- Shipped:
fd1ea4707(feature),8c3d6e7bb(test + cross-account adoption fix) - Why it's here:
format.test.ts(10) +roundtrip.test.ts(6) cover the pipeline with fake-indexeddb + pass-through crypto. Unit coverage is blind to: real AES-GCM through Web Crypto at browser speed, the File-download + re-upload path, Blob memory behaviour at multi-MB sizes, the passphrase-prompt modal flow, and the MyData settings-section wiring. - Steps:
pnpm run mana:dev. Log in, create a handful of notes + tasks + events.- Settings → Meine Daten → export (full, no passphrase). Download the
.manafile. - Inspect the archive (
unzip -l mana-full-*.mana) — expectmanifest.json+data/*.jsonl. - DevTools → Application → Storage → "Clear site data" on
localhost:5173. Reload. Log in again. - Settings → Meine Daten → import → pick the
.manafile. Check progress bar ticks, summary lists tables + row counts. - Open each module, confirm rows are back and readable (encrypted fields decrypt cleanly).
- Repeat 1–6 with passphrase on — export with passphrase, expect
data.sealedentry in the zip, re-import must prompt for the passphrase. Try once with the wrong passphrase, expect the friendly "Passphrase stimmt nicht" message, not a crash. - Cross-account: export from account A, log out, log in as account B (or a fresh signup), import. Confirm rows adopted by B (spaceId starts with
_personal:<B-userId>, not A's). Verify in DevTools → IndexedDB.
Shared Space — two-user smoke test
- Priority: 🟠 important
- Shipped: spaces-foundation 15-commit block (2026-04-20, see
project_spaces_foundation.mdmemory) - Why it's here: the data-layer + RLS + invite-flow integration can only be verified with two real sessions across two browser profiles.
- Steps: full walkthrough at
docs/plans/shared-space-smoketest.md— create Family Space, invite, accept, verify cross-user sync on calendar + recipes.
Articles bookmarklet — consent-walled sites
- Priority: 🟡 nice to have
- Shipped: articles M1–M9 (see
project_news_research_module.md/ plan atdocs/plans/articles-module.md) - Why it's here: the Weg-2 browser-HTML bookmarklet was built to route around server-side fetchers that hit a consent dialog (Golem, Spiegel, Zeit). No automation can exercise the real consent-wall detection path.
- Steps:
- Install the bookmarklet (Settings → Articles, drag to bookmarks bar).
- Navigate to a known consent-walled article (golem.de is the reference case — the original bug report).
- Accept the consent banner in the source tab, then click the bookmarklet.
- Expect a new tab that auto-saves the article; no extra "save to Leseliste" click required.
- Verify the saved article has readable body + title in
/articles.
MCP gateway + Persona-runner — end-to-end live smoke
- Priority: 🟠 important
- Shipped:
16c881833(M1+M1.5 MCP gateway),493db0c3b(M2.a-c persona schemas + seed),f07eae3c0(M3.b-d tick loop),eb8fac23e(tool_use_id pairing + audit),5a5e24f58(docker searxng fix). Plan atdocs/plans/mana-mcp-and-personas.md. Memory:project_mana_mcp_personas.md. - Why it's here: ~2600 lines of service code, 14 automated tests passed (type-check × 4, svelte-check, AES round-trip, HMAC 3-way parity, tool-registry integrity, seed dry-run, boot smokes × 2, Playwright config parse, drizzle SQL generate, vitest 21/21), but none of it has run against a live Postgres + mana-auth + Anthropic. Unit tests are blind to: real JWT issuance + SSO cookie flow, mana-sync wire-format mismatches, Dexie-table-name case drift, Better-Auth org-list response shape, Claude Agent SDK streaming edge-cases, encryption MK unwrap through the real vault endpoint, ZK-user rejection path.
- Steps:
pnpm dev:mana:all— brings up Postgres + Redis + MinIO + searxng + all dev servers.cd services/mana-auth && bun run db:push(or verify migrations 005 + 006 already applied).pnpm setup:dev-user— createstills95@gmail.comas founder. Log in via the web app, copy the JWT from DevTools → Application → Cookies →better-auth.session_token(or use the API):export MANA_ADMIN_JWT=$(curl -s -X POST localhost:3001/api/v1/auth/login \ -H 'content-type: application/json' \ -d '{"email":"tills95@gmail.com","password":"Aa-123456789"}' | jq -r .token)export PERSONA_SEED_SECRET=$(openssl rand -hex 32)— same value must stay in use for the runner + visual suite.pnpm seed:personas— expect "✓ Done. 10 personas upserted." Confirm in Postgres:
Row count should be exactly 10, all founder/persona.psql $DATABASE_URL -c "SELECT email, kind, access_tier FROM auth.users WHERE kind='persona'"pnpm --filter @mana/mcp-service dev— tail the log for[mana-mcp] listening on :3069.- Quick Claude-Code MCP check (optional): drop
.mcp.jsonwith{"mcpServers":{"mana":{"type":"http","url":"http://localhost:3069/mcp","headers":{"Authorization":"Bearer $YOUR_JWT","X-Mana-Space":"$YOUR_SPACE_ID"}}}}and/mcp listin Claude Code — expect 13 tools. - Start the runner:
Expectexport MANA_SERVICE_KEY=$(grep MANA_SERVICE_KEY .env.development | cut -d= -f2) export ANTHROPIC_API_KEY=sk-ant-... # your real key pnpm --filter @mana/persona-runner dev[mana-persona-runner] listening on :3070without any "MANA_SERVICE_KEY missing" or "ANTHROPIC_API_KEY missing" warning. - Fire one tick manually:
Expectcurl -s -X POST localhost:3070/diag/tick | jq{ok: true, result: {due: 10, ranSuccessfully: N, failed: [...], durationMs: …}}. Anyfailed[]entries with error messages are the actual bugs to chase. - Inspect what landed:
Rows should include calls topsql $DATABASE_URL -c "SELECT persona_id, tool_name, result, latency_ms FROM auth.persona_actions ORDER BY created_at DESC LIMIT 30" psql $DATABASE_URL -c "SELECT persona_id, module, rating, notes FROM auth.persona_feedback ORDER BY created_at DESC LIMIT 30"habits.create,todo.create,notes.create,journal.addetc., plus 1–5 ratings per module a persona used. - Encryption check — encrypted fields must be base64+
enc:1:prefixed in the wire table but decrypt cleanly client-side:
Expect strings beginning withpsql $MANA_SYNC_URL -c "SELECT data->'title' FROM sync_changes WHERE table_name='tasks' AND actor->>'kind'='persona' LIMIT 3""enc:1:". Then log in as that persona in the web app — the tasks should show plaintext titles.
Persona visual regression — capture first baselines
- Priority: 🟡 nice to have
- Shipped:
79d112657(M5.a scaffold). One flow (home.spec.ts), two viewports (desktop + Pixel 5). Extension pattern documented intests/personas/README.md. - Why it's here: Playwright config parses and lists tests, but no baseline screenshot has ever been captured. First capture must be a human eyeballing — otherwise the first CI run locks in whatever weird transient state happened at t=0 as "correct".
- Steps:
- Complete the MCP/persona-runner smoke above so each persona has real content (habits, tasks, journal entries). Without content, baselines are meaningless — every module is empty for everyone.
- Keep the web app running (
pnpm mana:devat :5173). PERSONA_SEED_SECRET=<same as seed> pnpm test:personas:update— writes PNGs undertests/personas/__snapshots__/home.spec.ts/home-adhd-student-{desktop,mobile}.png.- Eyeball each generated PNG before committing. Look for: persona's actual name in header, any unexpected "no data" empty states that mean step 1 content didn't land, layout breakage on mobile.
git add tests/personas/__snapshots__/ && git commit -m "test(personas): baseline home-tour for Anna"— now CI has a reference. Copyhome.spec.tstotodo.spec.tsetc. to add more flows; repeat step 3–5 per flow.
Articles — PWA share-target
- Priority: 🟡 nice to have
- Shipped: share-target configured in
vite.config.ts(routes to/articles/addwith?url+?text+?title) - Why it's here: only works on an installed PWA; dev-mode SW is disabled.
- Steps:
pnpm build && pnpm preview, install the PWA (Chrome "install Mana" prompt on desktop, or Android "add to home screen").- Share a URL from the OS share sheet (Chrome Android, WhatsApp, Mail) → expect Mana to appear and land on
/articles/addwith the URL pre-filled in the form.
Recently closed
(Move entries here with the date they were verified, prune after one release.)
None yet — this file was introduced 2026-04-22.
How to add an entry
When you ship a feature with unit tests but no browser / device click-through, add a new ### <feature> section under "Open" using the format above. Keep steps reproducible: assume a fresh clone and pnpm docker:up && pnpm run mana:dev. If the steps cross 20 lines, split them into a dedicated walkthrough file under docs/plans/ and link from here.