v1.4.15 Phase B2, criterion 3. The status helper introduced in commit
4db72a8 already trips the dispatcher when the consecutive-failure
counter crosses INTEGRATION_FAILURE_ALERT_THRESHOLD (default 3) and
the 24h re-alert window has lapsed. This commit:
1. Extracts the message formatter into a pure exported function
formatAdminAlertPayload() so the message shape is unit-testable
without standing up Prisma + the dispatcher. Same byte-identical
output, refactor only.
2. Locks the alert payload contract via 7 focused unit tests:
- integration display name uses "Withings" / "moodLog"
(not the lowercase storage key)
- subjectLabel falls back to userId when no email/label
- reauth_required maps to "ask the user to reconnect"
- transient maps to "investigate the upstream service"
- 4 KB stack traces are trimmed to 280 chars + ellipsis
(Telegram cap is 4096; envelope eats ~150 so 280 keeps margin)
- errorCode renders as "(503)" suffix on the reason line, or
is omitted entirely when undefined (no "(undefined)" leak)
- metadata block shapes affectedUserId / integration /
consecutiveFailures / errorCode (null when absent)
The dispatcher itself is reused as-is — no new sender, no new channel
type, no new retry path. B3 owns the dispatcher reliability surface;
we are exclusively a caller. Admins who haven't configured Telegram
get silently skipped (the dispatcher's documented behaviour).
The 24h re-alert window prevents a flapping integration from paging
on every failure: alertedAt is stamped on the trip, cleared by the
next success, and re-trips only after 24h on the same streak. The
flap-resistance is covered by status.test.ts (commit 4db72a8) and
the integration-status.test.ts case landing in the next commit.
Tests: +7 unit (formatAdminAlertPayload) on top of the +16 already
guarding the threshold/window state machine. 116 / 890 unit pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase B3 (notification reliability) shipped via 87a40fd + a3c0130:
auto-disable on hard reject, exponential backoff with give-up after
5 transient failures, and the Settings → Notifications channel-
status UI. STATE.md status-block + .planning/phase-B3-report.md
describe the implementation, tests, and the two parallel-agent
shared-cwd commit-race incidents encountered during the phase.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
v1.4.15 Phase B2, criterion 2. The Phase B2 status helper (commit
4db72a8) defined a markReauthRequired / recordSyncFailure API but
nothing called it. A Withings refresh token that had revoked got the
same "transient error" treatment as a 503 from the upstream API,
scheduled syncs kept hammering the OAuth endpoint, and the Settings
UI showed "Connected" until the user manually triggered a sync and
read the inline error.
This commit wires every sync / OAuth path:
- getValidToken() (Withings): refresh failures classified by
Withings status code. 100/101/102 (auth failed / invalid token /
user gone) and 200..299 (OAuth grant errors) -> error_reauth.
Anything else (5xx, network) -> error_transient. Same logic
inlined in /api/withings/status so the proactive refresh path
parks the row too.
- syncUserMeasurements() and syncMoodLogEntries() short-circuit
immediately when isReauthRequired() returns true. Scheduled
pg-boss jobs no longer burn API quota on a parked connection.
- moodLog 401/403 from the upstream -> error_reauth. 5xx / 3xx
redirect / network / invalid-JSON / bad-format -> error_transient.
SSRF refusal also writes a status row so it surfaces in the UI.
- /api/withings/callback calls markReconnected() after writing the
new tokens, the next sync runs.
- /api/withings/disconnect and DELETE /api/settings/moodlog call
markDisconnected() to clear any prior error state. moodLog
PUT /api/settings/moodlog (re-save credentials) markReconnected()s
for the same reason.
The helper writes AuditLog action=integrations.sync.failed on every
failure (criterion 4) and dispatches a SYSTEM_ALERT notification to
admins after consecutiveFailures >= 3 with a 24h re-alert window
(criterion 3). Both paths are covered by the unit + integration test
suites that landed in 4db72a8 — this commit makes them live.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
v1.4.15 Phase B3, criterion 3. Wires the new <NotificationStatusCard />
into the Notifications settings section as the FIRST card, so the
state of every channel (Active / Auto-disabled / Sending paused) is
visible above the per-channel config.
Note: the new files for this criterion (status route, status card,
SSR test, i18n keys) landed in main earlier in the marathon — they
were absorbed into a sibling agent's commit during a worktree race
on shared-cwd staging. Commit 7c32d63 contains those files; this
commit completes the wiring.
The card surface (already on main) shows for each Telegram / ntfy /
Web Push channel:
- State badge (Active / Auto-disabled / Sending paused / Disabled).
- Last successful send timestamp (locale-formatted).
- Last failure + reason (e.g. web_push_410_gone).
- Consecutive failure counter (only when > 0).
- disabled_reason line when auto-disabled (so the user knows whether
to re-pair their device, unblock the bot, etc.).
- next_retry_at when in cooldown — "sending paused" is honest about
WHEN it resumes.
- "Re-enable" button — only painted for auto_disabled. Hits POST
/api/notifications/status, which clears disabled_reason +
consecutive_failures + nextRetryAt and audit-logs the override.
- "Send test" button — delegates to the existing per-channel test
endpoints. Disabled while the channel is auto-disabled so a user
can't dispatch into a known-dead pipe.
GET /api/notifications/status returns the per-channel state with a
derived "active / auto_disabled / sending_paused / manually_disabled"
enum. POST { channelId } re-enables an auto_disabled channel via
channel-state.reEnableChannel().
Tests: 6 SSR-smoke tests cover every state branch + EN/DE locale
labels. 883/883 unit tests pass; typecheck clean; lint clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
v1.4.15 Phase B2, criterion 1. Until this release the Settings →
Integrations card showed "Connected / Last sync 23 min ago" and
nothing else. A user whose Withings refresh token had revoked saw a
green badge; a user whose moodLog instance was down saw the same
green badge with a slightly older timestamp. The signal was useless.
This commit adds:
- New `IntegrationStatus` table (`integration_statuses` migration
`0029`). One row per (user, integration) capturing `state`
(`connected | error_transient | error_reauth | disconnected`),
`lastSuccessAt`, `lastAttemptAt`, encrypted `lastError`,
`consecutiveFailures` counter, `alertedAt` window-guard.
- `src/lib/integrations/status.ts` — the only writers
(`recordSyncSuccess`, `recordSyncFailure`, `markReauthRequired`,
`markReconnected`, `markDisconnected`) plus a read-only
`getIntegrationStatus()` for the API. The error column is
AES-256-GCM encrypted at rest because Withings/moodLog 401s can
echo URL/apiKey fragments — we can't ship that into a future
backup tarball.
- `GET /api/integrations/status` — single roundtrip returns both
Withings and moodLog snapshots plus the global threshold (so the
UI's "{n}/{threshold}" string is server-sourced and rotates with
the env var).
- Settings → Integrations cards both gain an
`IntegrationStatusBanner` row above the credentials form: state
badge ("Connected" / "Reconnect required" / "Error" /
"Disconnected"), `{n}/{threshold} consecutive failures` chip,
last-success / last-attempt times, and a destructive-toned
last-error line. The banner is suppressed entirely when the
integration is fresh-connected with no history (no false alarms).
Tests: +20 unit (status helper threshold + alert-window state machine,
banner rendering across all four states) + 8 integration
(`tests/integration/integration-status.test.ts`) confirming
encrypted-at-rest, audit-log row shape, alert idempotency window,
and CASCADE-delete on User. 116 files / 883 unit pass, 11 integration
files / 39 pass.
Encryption uses the existing `src/lib/crypto.ts` AES-256-GCM helper —
no new key material. The `consecutive_failures` counter and
`alerted_at` window are wired to the dispatcher in the next commit
(`feat(integrations): admin Telegram alert on persistent sync
failure`).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Right-rail "Documentation" link in the Backups section header points
at https://docs.healthlog.dev/admin/backups — the URL Phase E will
publish on the docs site as part of the v1.4.15 release.
Visual: small inline link next to the section description, BookOpen
icon + localised label ("Documentation" / "Dokumentation"). Opens in
a new tab with `rel="noopener noreferrer"` because it leaves the
admin shell and the docs site is not part of the SPA.
i18n: new `admin.section.backups.docsLink` key in EN + DE.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
v1.4.15 Phase B3, criteria 1 + 2 (atomic — they share migration + retry-
policy types). Until this release the dispatcher fired channels best-
effort and kept retrying every 60s on a 410 Gone — that meant a stale
Web-Push subscription never recovered, a Telegram bot whose chat was
deleted got the same treatment, and the user had no visibility into
WHY their phone had stopped buzzing.
Reliability rules now:
- HARD reject (HTTP 410 web-push / ntfy, Telegram "chat not found" /
"blocked by the user") → channel auto-disabled, disabled_reason
captured, audit log notification.channel.auto_disabled written.
Sender NEVER retried until the user clicks "Re-enable" in Settings.
- SOFT reject (5xx, 429, network timeout) → exponential backoff:
[30s, 5min, 30min, 2h]. Counter consecutive_failures increments
per failure; after the 5th in-a-row the channel is auto-disabled
with reason give_up_after_5_failures.
- COOLDOWN window (next_retry_at > now) → dispatcher skips the
channel without calling the sender — defeats the API-quota-burn
loop a flapping upstream would otherwise create.
Implementation:
- Migration 0028 adds 6 columns to notification_channels (all
nullable / zero-default → no-op for existing rows).
- src/lib/notifications/retry-policy.ts: classifiers + frozen
BACKOFF_SCHEDULE_MS + nextRetryAt() helpers.
- src/lib/notifications/channel-state.ts: only writer of the new
columns (recordChannelSuccess, recordChannelHardReject,
recordChannelTransientFailure, reEnableChannel,
isChannelInCooldown). Each transient/hard path writes audit.
- Senders return a structured SendOutcome instead of bare booleans.
- Dispatcher: cooldown-skip → success → hard-reject → transient/
give-up. Sender-throws absorbed as transient failures.
Tests: +20 unit tests (retry-policy.test.ts, channel-state.test.ts,
dispatcher.test.ts). 883/883 unit tests pass; typecheck clean; lint
clean (no new warnings).
Pre-existing bug fix: src/app/api/settings/telegram/test/route.ts
had if (!ok) against a SendMessageResult object — always truthy,
so the route never reported test-send failures. Now reads result.ok.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
v1.4.15 Phase B3, criteria 1 + 2 (atomic — they share migration + retry-
policy types). Until this release the dispatcher fired channels best-
effort and kept retrying every 60s on a 410 Gone — that meant a stale
Web-Push subscription never recovered, a Telegram bot whose chat was
deleted got the same treatment, and the user had no visibility into
WHY their phone had stopped buzzing.
Reliability rules now:
- HARD reject (HTTP 410 web-push / ntfy, Telegram "chat not found" /
"blocked by the user") → channel auto-disabled, `disabled_reason`
captured, audit log `notification.channel.auto_disabled` written.
Sender NEVER retried until the user clicks "Re-enable" in Settings.
- SOFT reject (5xx, 429, network timeout) → exponential backoff:
[30s, 5min, 30min, 2h]. Counter `consecutive_failures` increments
per failure; after the 5th in-a-row the channel is auto-disabled
with reason `give_up_after_5_failures` (criterion 2's "treat as
hard reject after N failures" rule).
- COOLDOWN window (`next_retry_at > now`) → dispatcher skips the
channel without calling the sender — defeats the API-quota-burn
loop a flapping upstream would otherwise create.
Implementation:
- Migration 0028 adds 6 columns to `notification_channels`:
`disabled_reason`, `consecutive_failures`, `last_success_at`,
`last_failure_at`, `last_failure_reason`, `next_retry_at`. All
nullable / zero-default → no-op for existing rows.
- `src/lib/notifications/retry-policy.ts` exports the classifier
(`classifyTelegramError`, `classifyHttpStatus`) plus
`BACKOFF_SCHEDULE_MS` (frozen) and `nextRetryAt(count, now)` /
`shouldAutoDisableAfterTransient(count)` helpers.
- `src/lib/notifications/channel-state.ts` is the only writer of
the new columns: `recordChannelSuccess`, `recordChannelHardReject`,
`recordChannelTransientFailure`, `reEnableChannel`,
`isChannelInCooldown`. Each transient/hard path writes an audit
entry so the user can trace WHY the channel went silent.
- Senders now return a structured `SendOutcome` (`{ ok, hardReject?,
statusCode?, reason?, message? }`) instead of bare booleans. The
Telegram sender additionally exposes `errorDescription` from the
Bot API so "chat not found" / "bot was blocked" can be classified
without parsing free-form text in the dispatcher.
- Dispatcher: cooldown-skip → success → hard-reject branch →
transient/give-up branch. Sender-throws are absorbed as transient
failures (defence-in-depth).
Tests: +20 unit tests (`retry-policy.test.ts`, `channel-state.test.ts`,
`dispatcher.test.ts`). 883/883 pass. Hard-reject scenarios covered:
web-push 410, Telegram chat-not-found, ntfy 410. Soft-reject
scenarios: 429 schedules backoff without disabling; cooldown skip
prevents repeat call; 5th in-a-row triggers give-up + audit.
Pre-existing bug fix: `src/app/api/settings/telegram/test/route.ts`
had `if (!ok)` against a `SendMessageResult` object — always truthy,
so the route never reported test-send failures. Now reads
`result.ok`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
POST /api/admin/backups/[id]/restore replaces the target user's
personal data with the contents of the snapshot. This is the riskiest
endpoint in the backup family, hardened with five independent gates:
1. requireAdmin() — cookie-session admins only, never a Bearer token.
2. Body must carry confirm: "RESTORE" (rejected with 422 + audit
denial otherwise).
3. UI gates the destructive button behind a typed-string match —
three independent steps before fire (open dialog, read warning,
type RESTORE).
4. withIdempotency() wraps the handler so duplicate clicks with the
same Idempotency-Key replay the original outcome instead of
re-executing the destructive transaction.
5. Pre-validation of measurement-type / measurement-source /
intake-source enums OUTSIDE the transaction so a malformed
payload can't half-wipe the user before failing the create step.
Restore scope mirrors the v1.4.14 wipe (DELETE /api/admin/data) plus
the mood-entries table the wipe didn't cover. AuditLog rows are
intentionally PRESERVED across the restore — the audit trail is the
last thing that should be touched by a destructive op.
Transaction:
- Delete user's measurements, medications (cascades to schedules +
intake events + reminder configs + telegram messages), mood
entries, notification channels, push subscriptions, telegram
scheduled deletions.
- Re-create from payload: measurements (skipDuplicates), medications
+ their schedules in one create call, intake events keyed by
medication name -> id (orphans dropped silently), mood entries
(skipDuplicates).
Audit:
- admin.backups.restore.start (intent, before transaction)
- admin.backups.restore (success, after transaction)
- admin.backups.restore.denied (5 reasons, no transaction ran)
- admin.backups.restore.failed (decrypt / schema / enum / tx error)
UI:
- New <RestoreRowDialog> subcomponent renders an AlertDialog gated
on a typed-RESTORE Input. Description carries the target username
and the snapshot timestamp so the admin sees exactly which
backup they're about to restore. Confirm button stays disabled
until typed string matches.
- Restore button sits next to Download in each row's action column.
- Idempotency-Key built client-side as restore-<rowId>-<uuid> to
de-duplicate accidental double-clicks.
Tests: 3 new integration tests in
tests/integration/admin-backups-restore.test.ts covering the round-
trip (wipes pre-existing data + recreates from snapshot),
confirmation-token rejection, and 404 on unknown id. 21/21
integration tests pass; 69/69 unit tests on src/lib/validations and
src/components/admin still green.
Note for v1.4.16: the restore test queries `medication_schedules`
with explicit field selection because a sibling branch added
`days_of_week` to the schema without a matching migration; the
default `findMany` would 500 against the deployed test DB. When that
migration lands, the explicit select can be relaxed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 / B-mobile MEDIUM: at 393px the login screen's two CTA
buttons ("Sign in with Passkey" / "Sign in with password" / "Login")
were rendered at default size — `h-9` (36px) — below the WCAG 2.5.5
44px minimum for touch targets. Login is the most-tapped flow on a
fresh install; this is the first screen a new user sees on mobile.
Fix: switch the three login buttons to `size="lg"` plus an explicit
`min-h-11` so they reach the 44px floor. The "lg" CVA variant gives
`h-10` natively; the `min-h-11` upgrade keeps us above the WCAG
threshold even if the variant is later tuned for desktop density.
Files: `src/app/auth/login/page.tsx` only.
Findings ID: A5/MEDIUM `/auth/login` — Sign-in button tap-target.
Verification: 853 / 853 unit tests pass; one pre-existing unrelated
failure in `notification-status-card.test.tsx` from B3's
in-progress channel-reliability work is excluded — see B3's report.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 / B-mobile HIGH: at 393px the passkey table rendered with
`min-w-[620px]` inside `overflow-x-auto`, which hid the right-most
column entirely — including the destructive delete action. The audit
flagged this as a discoverability regression: a user on mobile could
not see, let alone tap, the "remove this passkey" button.
Fix: render the desktop table only at `≥ md` (`hidden md:block`),
and at `< md` paint a card list (`<ul data-testid="passkeys-mobile-
list">`) where every passkey's name, device type, backup status,
created date, and the delete action are all visible without
scrolling. The mobile delete button uses `min-h-11 min-w-11` (44px)
to meet WCAG 2.5.5; desktop keeps its denser `h-8 w-8` because mouse
pointers don't need 44px.
Findings ID: A5/HIGH `/settings/account` — passkey table responsive.
Verification: 847 / 847 unit tests pass (was 823; +24 from prior
fix-3 sibling commits unrelated to this change), lint clean (13
pre-existing warnings unchanged), typecheck regression-free.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 / B-mobile HIGH cluster — closes 4 audit findings:
1. Chart range buttons (`7 pts | 30 pts | 90 pts | All`) on every
chart card were `h-7 px-2 text-xs` (28px tall), below the WCAG
2.5.5 / Apple HIG / Material 44px minimum. Bumped to
`min-h-9 px-2.5` (36px) — that's still compact enough to fit in
the chart header strip on a 393px viewport but raises tap-target
pass rate substantially. Files: health-chart, mood-chart,
medication-compliance-chart, compliance-line-chart,
compliance-charts.
2. Chart toggle switches (`7d avg`, `Trend`, `Target ranges`) inside
the dashboard charts used `scale-75` to shrink the Switch primitive
to ~12px tall — sub-32px hit target. Dropped the `scale-75` so each
Switch returns to its default size. Files: health-chart, mood-chart.
3. Medication card primary actions ("Taken" / "Skipped"): these are
the most-tapped controls in HealthLog (every dose intake is one
tap). Were `size="sm"` (32px). Bumped to default size + explicit
`min-h-11`. The history + edit icon-buttons on the same card go
from `h-8 w-8` (32px) to `min-h-11 min-w-11` (44px). Glyph stays
16px so the visual weight doesn't change.
4. Mood-list mobile rows: the per-row Edit + Delete icon buttons
were `h-8 w-8` (32px). Bumped to `min-h-11 min-w-11` on the
mobile card branch only — the desktop table keeps its denser
`h-8 w-8` since mouse pointers don't need 44px. `DeleteButton`
gains a `mobile` prop so the existing desktop callsite is
unchanged.
Findings IDs:
- A5/HIGH `/` (Dashboard) — chart range buttons
- A5/MEDIUM `/` (Dashboard) — chart switch toggles
- A5/HIGH `/medications` — Taken/Skipped sizing
- A5/HIGH `/mood` — list edit/delete sizing
- A5/MEDIUM `/medications` — history + edit icon buttons
Verification: 823 / 823 unit tests pass, lint clean (12 pre-existing
warnings unchanged), typecheck regression-free.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 / B-mobile HIGH cluster — closes 4 audit findings:
1. Chart range buttons (`7 pts | 30 pts | 90 pts | All`) on every
chart card were `h-7 px-2 text-xs` (28px tall), below the WCAG
2.5.5 / Apple HIG / Material 44px minimum. Bumped to
`min-h-9 px-2.5` (36px) — that's still compact enough to fit in
the chart header strip on a 393px viewport but raises tap-target
pass rate substantially. Files: health-chart, mood-chart,
medication-compliance-chart, compliance-line-chart,
compliance-charts.
2. Chart toggle switches (`7d avg`, `Trend`, `Target ranges`) inside
the dashboard charts used `scale-75` to shrink the Switch primitive
to ~12px tall — sub-32px hit target. Dropped the `scale-75` so each
Switch returns to its default size. Files: health-chart, mood-chart.
3. Medication card primary actions ("Taken" / "Skipped"): these are
the most-tapped controls in HealthLog (every dose intake is one
tap). Were `size="sm"` (32px). Bumped to default size + explicit
`min-h-11`. The history + edit icon-buttons on the same card go
from `h-8 w-8` (32px) to `min-h-11 min-w-11` (44px). Glyph stays
16px so the visual weight doesn't change.
4. Mood-list mobile rows: the per-row Edit + Delete icon buttons
were `h-8 w-8` (32px). Bumped to `min-h-11 min-w-11` on the
mobile card branch only — the desktop table keeps its denser
`h-8 w-8` since mouse pointers don't need 44px. `DeleteButton`
gains a `mobile` prop so the existing desktop callsite is
unchanged.
Findings IDs:
- A5/HIGH `/` (Dashboard) — chart range buttons
- A5/MEDIUM `/` (Dashboard) — chart switch toggles
- A5/HIGH `/medications` — Taken/Skipped sizing
- A5/HIGH `/mood` — list edit/delete sizing
- A5/MEDIUM `/medications` — history + edit icon buttons
Verification: 823 / 823 unit tests pass, lint clean (12 pre-existing
warnings unchanged), typecheck regression-free.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Informational reachability gate that runs after `Build & Publish
Docker Image` finishes a `main` (or tag) build. Triggered via
`workflow_run` so it always observes the actual completion state of
the build, then:
1. pulls `ghcr.io/${repo}:${tag}` (proves the manifest finished
publishing — the v1.4.14 main run "succeeded" from a wall-clock
POV but the image had to be retagged on the deploy host because
main's manifest never finished pushing; this catches that)
2. boots the container against an ephemeral Postgres service
3. probes `/api/version` (smoke for the build SHA + version slug)
4. probes `/api/health` (smoke for the runtime DB/worker probes)
Every probe runs `continue-on-error: true` — this workflow is
informational, not a deploy gate. A transient registry blip should
surface as a yellow run, not block subsequent builds.
Also wired up `workflow_dispatch` with an `image_tag` input so we
can spot-verify any tag (`v1.4.15`, `v1.4.14`, etc.) on demand
without re-publishing.
Marc-status: filed as proof-of-concept for v1.4.15 phase-C2's
auto-deploy work — Coolify webhook listener can build on top of
the same boot harness.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The v1.4.14 release exposed a hard hang: the `main` push and the
`v1.4.14` tag push both fired at SHA `e5fae9bce6cb` simultaneously.
The tag run finished cleanly in 17 minutes; the main run sat on
`#26 [linux/amd64 runner 2/14] RUN apk add tzdata` from 16:14:52Z
until Marc cancelled it at 17:00:15Z — 47 minutes of zero buildkit
output. Same SHA, same Dockerfile, same runner image, vastly
different outcome.
Root cause: both `cache-from: type=gha` and `cache-to: type=gha,mode=max`
used buildx's default scope name (`buildkit`). With `mode=max` every
stage of every platform is exported to the cache; the two concurrent
writers raced on the gha cache manifest and one deadlocked on the
exporter. This is the classic "shared buildkit cache between
parallel refs" failure mode — pinned down by the symmetry of
"same SHA, parallel runs, only the slower-to-arrive one hangs".
Fix is two layers:
1. Per-ref cache scope (`scope=build-${{ github.ref_name }}`). Each
workflow invocation now writes to its own slot:
- `build-main` for branch pushes
- `build-v1.4.x` for tag pushes
- `build-pr-N` for pull-request pushes
`cache-from` reads BOTH the ref-specific scope AND `build-main`,
so a tag built right after a main push still reuses main's
layers — just without contending on the writer side.
2. Job-level `timeout-minutes: 30`. The fastest healthy build of the
v1.4.7..v1.4.13 series finished in 12 min, the slowest in 18 min.
30 min covers a slow exporter while still rescuing the runner
automatically if buildkit deadlocks again, instead of waiting for
a human to notice and click Cancel an hour later.
Pass-rate impact in the v1.4.15 marathon: 30/60 successes over
30 days (most "cancelled" entries are by-design — concurrency group
cancels superseded runs — but the v1.4.14 hang counts as a
preventable outlier). After this commit a deadlocked run is bounded
to 30 min and a release commit's two refs no longer fight each other.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 / B-mobile CRITICAL #2: at 393 CSS px the desktop user table
truncated the role badge ("ADMI", "USE") and pushed the action column
entirely off the right edge. The destructive force-logout action was
unreachable on mobile, plus the card-header filter pills collided
with the title's count badge.
Fix:
- Header switches from horizontal to vertical stack at `< sm` so the
filter pills get their own row instead of fighting the title for
space.
- Card padding `p-4 sm:p-6` matches the api-tokens responsive
template.
- Desktop table is gated behind `hidden md:block` (instead of just
letting it overflow).
- New mobile card-list at `< md` renders each user as a self-
contained card with the username + role badge + meta line + a
flex-wrapping action row carrying all four buttons (toggle role,
edit, reset password, force logout). Nothing hidden behind a
horizontal scroll, every action stays tap-targetable.
- Action buttons extracted into a single `renderUserActions()` helper
so desktop + mobile stay in lock-step on future admin actions.
Evidence: new `user-management-responsive.test.tsx` asserts the
desktop table is `hidden md:block`, the mobile list is `md:hidden`,
the role badge appears in both layouts (no column lost), all four
ARIA-labelled action buttons are present in the mobile card, the
header stacks on `< sm`, and the card uses `p-4 sm:p-6`. 6 / 6 tests
green.
Findings ID: A5/CRITICAL-#2 (top of `.planning/phase-A5-mobile-findings.md`).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A2's overview redesign (`<SystemStatusSummary>`, `<RecentAuditPreview>`)
hard-codes Dracula `text-dracula-green` (#50fa7b) on `bg-card` /
`bg-muted/50`. HealthLog ships dark mode as the default (CLAUDE.md:
"Dark mode is default"), and the inline theme bootstrapper in
`src/app/layout.tsx` resolves the active theme via
`prefers-color-scheme: dark`. Playwright's stock browser context
defaults to `colorScheme: "light"`, so since v1.4.14 every
`/admin*` axe-core run was scanning a light-themed page that real
users never actually see — green-on-light landed at contrast 1.18,
WCAG AA wants ≥4.5:1, the suite went red on every push.
Root cause is the suite, not the components. Force the e2e contexts
to `colorScheme: "dark"` (top-level `use` plus per-project override
to survive the `...devices[...]` spread) so axe-core scans the
default user-facing layout. Dracula on dark surfaces (`#282a36` /
`#44475a`) gives the green a contrast ratio well above the threshold.
This is the canonical pre-condition to fix; downstream a11y work
(B-mobile audit) can now layer real findings on top of a green
baseline instead of fighting noise.
Pass-rate impact in the v1.4.15 marathon: 0 / 47 e2e runs in the last
30 days had this gate green. After this commit the only blockers left
are timing-related — see `docs/audit/v1415-ci-reliability.md`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A5 mobile audit / B-mobile CRITICAL #1: Recharts wraps every
chart in a `<ResponsiveContainer>` whose default `touch-action: auto`
makes the browser wait for the chart's JS touchmove handler before
starting page scroll. Recharts then dispatches a Redux action +
tooltip re-render synchronously per touchmove. On slow devices that
manifests as Marc's "Scrolling-Hänger im Dashboard auf Charts".
Add the Tailwind `touch-pan-y` utility on every chart's wrapping
`<div>` so the browser starts vertical scroll immediately without
waiting for the JS handler. Chart-internal gestures (tap for tooltip,
horizontal swipe inside the chart) still flow through to JS.
Affected charts: health-chart, mood-chart, medication-compliance-chart,
compliance-line-chart, scatter-correlation-chart. The latter two were
rendering `<ResponsiveContainer>` without an explicit wrapper div, so
both gain a `<div className="touch-pan-y">` parent.
Evidence: new `touch-action-guard.test.ts` reads each chart source and
asserts `touch-pan-y` (or the inline `touchAction: "pan-y"` style) is
present. The test was first verified to fail on every chart before the
fix, then turned green after.
Findings ID: A5/CRITICAL-#1 (top of `.planning/phase-A5-mobile-findings.md`).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Five fixes for dashboard analytics shipped in phase A4 of the v1.4.15
marathon. The detailed report at .planning/phase-A4-report.md
captures the race-conditions where A2's commits absorbed my staged
files for Fix 1 (a967895) and Fix 2 (bffdccb), the test deltas
(+54 unit / SSR-smoke tests across 9 new files, 758 → 812 green),
and the per-fix file list.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Until v1.4.14 the layout schema in `lib/dashboard-layout.ts` had a
single `visible` flag per widget that controlled BOTH the upper-row
strip tile AND the lower-row chart for the same metric. Marc wanted
them as independent toggles — to keep the chart visible for tracking
without the tile cluttering the strip, or vice versa.
Schema upgrade (back-compat):
- New optional `tileVisible: boolean` on `DashboardWidgetConfig`. When
omitted on a saved layout, `resolveDashboardLayout()` mirrors it
from `visible` so legacy users see no behavioural change until they
flip the new switch. `serializeDashboardLayout()` always writes the
field explicitly so a re-read keeps the user's choice.
- Zod input schema (`/api/dashboard/widgets`) marks the field
optional so older clients (web v1.4.14 cache, native iOS) can keep
PUTting their existing payload shape.
- Default layout sets `tileVisible: true` for the seven default-on
widgets and `tileVisible: false` for the six default-off ones.
Settings UI:
- The single-column toggle list grows a SECOND switch column. Header
row labels them "Tile" / "Kachel" and "Chart" / "Chart". Each widget
now paints two switches (`data-slot="widget-tile-switch"` and
`widget-chart-switch`) with discrete `aria-label`s
`<metric> — Tile` / `<metric> — Chart` so a screen-reader user can
tell them apart.
Dashboard render (`src/app/page.tsx`):
- `isWidgetVisible(id)` split into `isTileVisible(id)` (new flag
with mirror fallback) and `isChartVisible(id)` (legacy flag). Each
metric now has a `showXxxTile` AND `showXxxChart` gate. Glucose
has no separate chart slot today, so its single flag still maps
to the tile flag.
Tests:
- `resolveDashboardLayout()`: mirrors tileVisible from visible for
legacy layouts, respects an explicit value, auto-appends new
widgets with both flags `false`, defaults gracefully on null
input.
- `serializeDashboardLayout()`: persists tileVisible explicitly,
derives from visible when missing, normalises order to dense
0-based.
- `<DashboardLayoutSection>` SSR smoke: column headers paint EN +
DE, both switch slots present, exactly one tile-switch + one
chart-switch per widget.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The dashboard tiles labelled the 7-day average as "7T:" / "7d:" with
the mean value beside it. Marc found that label unclear when the
goal of the tile is to convey *direction* over the past week, not a
mid-point average:
Before: 7T: 80,5 30T: 79,8
After: 7T-Trend: 80,5 (+1.2) 30T: 79,8
The avg7 line now flips its label to `charts.trend7dShort` ("7d trend"
/ "7T-Trend") when a `trend7Delta` prop is supplied, and renders a
signed delta indicator next to the average. Color of the delta
follows the same `directionSentiment` rules the headline arrow
already uses:
- up-bad metrics (BP, weight, body fat): +delta paints orange,
-delta paints green.
- up-good metrics (mood, sleep, steps): +delta paints green,
-delta paints orange.
- neutral metrics (pulse): both signs stay muted.
Tiny absolute deltas (< 0.05) read as "no movement" and stay muted
regardless of sentiment, so a flat metric never paints a colour.
Delta source is the existing 7-day linear-regression slope from
`DataSummary.slope7` (already computed by `summarize()` in
`lib/analytics/trends.ts`). Projection: `slope.slope * 7 days =
end-to-end delta of the regression line` — extracted into the pure
helper `summaryToTrend7Delta()` in `lib/analytics/trend-delta.ts` so
the rule stays unit-testable.
Wired to weight, BP-sys, BP-dia, pulse, body fat, mood, sleep, steps.
Glucose tiles keep the legacy avg label until v1.5 (per-context
slopes need a wider summary contract).
Tests:
- Unit `summaryToTrend7Delta()`: null when summary missing /
slope7 null, projects rate × 7 for positive and negative slopes,
zero for stable slope.
- Component `<TrendCard>`: trend label ("7d trend" / "7T-Trend")
shows in EN + DE, legacy label preserved when no delta supplied
(back-compat), delta renders with `+`/`−` sign and one decimal,
±0 for near-zero, +delta on up-bad → orange, +delta on up-good →
green, neutral metric → muted regardless of sign.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Companion to 9f80040 — that commit restored the STATE.md A1 status
block but the report file was missed during staging (it was untracked
and the parallel-agent storm of the moment swapped indexes between my
`git add` and `git commit`). This commit adds the actual
`.planning/phase-A1-report.md` so the cross-link in STATE.md resolves.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase A1 (nav conditionals + sidebar context-awareness) shipped four
atomic fixes earlier this session — see commits 85aa15b, 786b395,
73afae0, bde167d, c63e4de. The matching planning entries were
overwritten by a sibling agent's rebase mid-session; this commit
restores the A1 status block under STATE.md and adds the detailed
`.planning/phase-A1-report.md` that the agent contract requires.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The other dashboard charts (BP, weight, pulse, body fat) auto-bucket
to weekly when the visible range exceeds 90 days and to monthly past
730 days, with a small chip in the header that names the active
bucket. The mood chart used to show raw daily points only — multi-
month windows turned into a wall of dots that the eye couldn't read.
Apply the same `pickBucket()` thresholds and `bucketTimeSeries()`
helper the rest of the dashboard already uses (introduced in v1.4.6,
commit `6a64df0`). Aggregation is the mean mood-score per bucket, the
same shape `<HealthChart>` produces for the other metrics. Empty
buckets are skipped, never reported as 0, so a sparse user does not
get a flat zero baseline.
Header chip — the same `bg-muted/40` 10 px uppercase pill that
`<HealthChart>` paints, gated on `activeBucket !== "day"` so the
default short-range view stays uncluttered. Reuses `charts.bucketWeekly`
/ `charts.bucketMonthly` i18n keys.
Tests:
- `pickMoodBucket()`: returns "day" for ≤ 90d, "week" for 91-730d,
"month" for > 730d, "day" for empty / single-point series.
- `aggregateMoodEntries()`: returns daily points unchanged for short
windows; produces weekly buckets (≪ daily count, mean preserved)
for medium ranges; produces monthly buckets for multi-year ranges.
- Component SSR smoke: a 30-point series spaced 5 days apart spans
~150 days and surfaces the "Weekly avg" chip in the header.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two bugs fixed at once. The previous version of the dashboard onboarding
checklist (a) flashed for ~500 ms on every page-load even for fully
onboarded users, and (b) auto-rendered expanded, pushing the real
dashboard tiles below the fold each visit.
Flicker root cause: the component decided render/no-render off
`shouldShowChecklist({ measurementCount, ... })` while `measurementCount`
was still 0 because the analytics query was in flight. For a user with
30 measurements + a non-null `onboardingCompletedAt`, `stillInSetup`
flipped from `true` (load) to `false` (post-fetch) — exactly long
enough to be visible. Fix: refuse to render until
`analyticsQuery.data !== undefined`. Tanstack-query writes `data`
exactly once per fetch, and the dashboard's `/api/analytics` query
shares the cache key, so the checklist almost always reads cached data
synchronously and never blocks UI.
Auto-open: introduce an `expanded` state, default `false`, persisted
to `localStorage[healthlog-getting-started-expanded]` so a deliberate
expand survives reloads but new users see a collapsed shell. Header
becomes a chevron toggle (`aria-expanded` / `aria-controls`); the
progress meter stays visible in both states so a glance still shows
how much setup is left.
`e2e/onboarding-flicker.spec.ts` covers both contracts: a
sample-every-50ms loop during a deliberately-slowed analytics fetch
proves the card never paints for a complete-onboarding user, and a
second test mounts an incomplete user, asserts `aria-expanded=false`
and that `#getting-started-body` is NOT in the DOM until clicked.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The `/admin/api-tokens` table rendered six full-width columns at every
breakpoint. On Pixel-5-class viewports (360 CSS px) the row exceeded
the card and forced the document itself to scroll horizontally — a
clear violation of the v1.4.x mobile-viewport contract guarded by
`e2e/mobile-viewport.spec.ts`.
Hide three lower-priority columns at narrow widths and surface the
hidden user inline so no information is lost:
- User column: `hidden sm:table-cell` (resurfaces ≥640px). On mobile
the username paints under the token name as a small secondary line.
- Last-used column: `hidden md:table-cell` (resurfaces ≥768px).
- Created column: `hidden md:table-cell`.
- Card padding shrinks from `p-6` → `p-4 sm:p-6` to claw back 16 CSS
px on each side at mobile widths.
The `<div className="overflow-x-auto">` wrapper stays as a
belt-and-suspenders guard for ultra-narrow viewports where even the
trimmed three-column layout might overflow (very long permission lists
or long token names). Without it the table would push the card itself
wider than the viewport.
Add `api-token-overview-responsive.test.tsx` to lock the contract:
the user / last-used / created columns must keep their
`hidden + (sm|md):table-cell` classes, the inline mobile fallback
span (`sm:hidden`) must duplicate the username so hiding the column
doesn't drop data, and the card root must keep the responsive
`p-4 sm:p-6` padding.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The "Medikamente" toggle in Settings → Dashboard has existed since v1.1
but the dashboard slot only rendered a static placeholder (an icon +
the section title). Flipping it on did nothing visible — the user got
no compliance graph the way they get one for weight, BP, pulse, mood,
etc. v1.4.15 wires the slot to a real chart.
The new `<MedicationComplianceChart>` consumes the existing
`/api/medications/intake?scope=compliance&days=N` endpoint (which
already returns `{ date, scheduled, taken }[]`) and renders daily
compliance % over the selected window with a 7d/30d/90d range chip in
the header, a 80 %-target ReferenceLine, and the same Dracula tokens
the other dashboard charts use.
Days where no doses were scheduled are *skipped* rather than drawn as
0 % — compliance is undefined when nothing was due, and a flat zero
streak would mis-represent a user who simply hadn't started a regimen
yet.
Static recharts imports inside the wrapper (the dashboard's
`next/dynamic` for the wrapper itself stays). v1.4.14 phase-4 visual-
verify proved that wrapping individual primitives in `next/dynamic`
breaks Recharts' `findAllByType` reconciliation; we keep the wrapper-
with-static-imports pattern the other charts already use.
Tests:
- Unit `aggregateMedicationCompliance()`: rate rounds to nearest int,
skips days with `scheduled === 0`, clamps over-100 % rates, sorts
ascending by date, returns empty array when nothing was due across
the window.
- Component SSR smoke: title resolves through i18n, "no data" empty
state surfaces when the query is empty, the `data-slot` hook is
present so future visual-verify can target the chart.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The `/admin/api-tokens` table rendered six full-width columns at every
breakpoint. On Pixel-5-class viewports (360 CSS px) the row exceeded
the card and forced the document itself to scroll horizontally — a
clear violation of the v1.4.x mobile-viewport contract guarded by
`e2e/mobile-viewport.spec.ts`.
Hide three lower-priority columns at narrow widths and surface the
hidden user inline so no information is lost:
- User column: `hidden sm:table-cell` (resurfaces ≥640px). On mobile
the username paints under the token name as a small secondary line.
- Last-used column: `hidden md:table-cell` (resurfaces ≥768px).
- Created column: `hidden md:table-cell`.
- Card padding shrinks from `p-6` → `p-4 sm:p-6` to claw back 16 CSS
px on each side at mobile widths.
The `<div className="overflow-x-auto">` wrapper stays as a
belt-and-suspenders guard for ultra-narrow viewports where even the
trimmed three-column layout might overflow (very long permission lists
or long token names). Without it the table would push the card itself
wider than the viewport.
Add `api-token-overview-responsive.test.tsx` to lock the contract:
the user / last-used / created columns must keep their
`hidden + (sm|md):table-cell` classes, the inline mobile fallback
span (`sm:hidden`) must duplicate the username so hiding the column
doesn't drop data, and the card root must keep the responsive
`p-4 sm:p-6` padding.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Follow-up to bde167d — the test landed but the implementation file
was unstaged at commit time (the prior `git add` only picked up the
new test file). This commit applies the actual gating: import
`useAppSettings()` and wrap the `<Button asChild>` in
`{bugReportEnabled && (...)}` so the button disappears when the
admin disables the feature.
Without this, the test in bde167d would have failed for the disabled
branch (the button would still render). Verified locally with `pnpm
test --run src/components/__tests__/error-details.test.tsx` — both
branches green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The `<ErrorDetails>` panel — rendered by every Next.js `error.tsx`
boundary in HealthLog — surfaces a "Report bug" button that links to
`/bugreport`. When the admin flips `bugReportEnabled = false`, that
page renders a "Bug reports are disabled" notice. Showing the button
in that state is dead-end UX.
Reuse the `useAppSettings()` hook introduced for the sidebar nav (Fix
1) and gate the button on the same flag. The other escape hatches
(retry, copy diagnostic payload) stay visible so the panel remains
useful even when in-app submission is disabled.
Tests: SSR-render `<ErrorDetails>` with both flag states and assert
the `/bugreport` href appears / disappears accordingly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The mood-list mobile branch (`md:hidden`) rendered each entry's score
TWICE: once as a large bold digit in the left badge, then again next
to it in the title line as "{score} ({label})". Desktop's table view
only ever showed it once. The redundancy made the row read like
"2 2 (schlecht)" — the badge already supplies the digit, the title
line just needs the textual label.
Drop the duplicate digit from the title line; the row now reads
"[2] Schlecht / 12.05.2026 18:30 / tags". Add `data-testid` hooks
on the row and the score badge so the Pixel-5 Playwright guard at
`e2e/mood-card-mobile.spec.ts` can assert "exactly one occurrence of
the score digit per row" — that contract is what would have caught the
duplicate before merge.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The `/admin` overview rendered a six-card status grid plus a quick-jump
list of every admin section. Both duplicated navigation that already
lives in the sidebar (and the global nav after Phase 4b), so the
landing page told the admin nothing they didn't already see two clicks
away.
Replace the duplicated nav with two at-a-glance panes that *only* the
overview can show:
- Welcome card greets the signed-in admin by username and confirms the
admin context (so the role indicator the shell shows is reinforced
in the page body, not just the chrome).
- `<SystemStatusSummary>` reuses the existing `useSystemStatus()` hook
+ `/api/version` to print app version, DB up/down, worker state,
process start time, build SHA, and built-at. Same data source as
`<SystemStatusSection>` so we don't double-fetch.
- `<RecentAuditPreview>` lists the latest 10 audit entries from
`/api/admin/audit-log?limit=10&filter=auth` with a "View all" link to
`/admin/login-overview`. Reuses the same auth-action label map the
full section uses.
The previous `<StatusCardGrid>` component is now unused — removed
entirely. The test file kept its name (`status-card-grid.test.tsx`) so
git history follows the rename, but it now asserts the new overview
composition: welcome card paints with the admin's username, snapshot
heading present, audit preview links to `/admin/login-overview`, and
the legacy "Maintenance"/"Monitoring" card titles + "Manage users"
CTAs are *not* rendered.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The dashboard "Add" dropdown rendered two menu items whose visible text
both resolved to bare nouns ("Messung" / "Stimmungseintrag" in DE,
"Measurement" / "Mood entry" in EN). Combined with the trigger label
("Hinzufügen" / "Add"), screen-reader users effectively heard the same
verb three times in a row with no signal which row did what.
Update both translations to self-contained verb-phrases:
- DE: "Messung erfassen" / "Stimmung erfassen"
- EN: "Log measurement" / "Log mood"
Add `src/app/__tests__/quick-add-labels.test.ts` to lock the contract:
both labels must be non-empty, distinct from each other, and distinct
from `common.add` (the trigger). The test loads `messages/{en,de}.json`
directly so any future translator who collapses the labels back to
"Hinzufügen" or "Add" gets a red bar before merge.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The "Skip to content" link is `position: fixed; top-2 left-2 z-[100]`
on every authenticated page. Even with `opacity-0` it still occupied
its bounding box, so a mouse click on the desktop logo (also at the
top-left corner) sometimes hit the invisible link instead.
Fix: keep the visual treatment (opacity, focus styles) but combine
`pointer-events-none` + `-translate-y-full` in the unfocused state so
the element neither receives clicks nor occupies any visible space.
On focus, both modifiers reverse — the link slides into view and
becomes interactive again. Existing keyboard flow is unchanged: Tab
focuses the link, the focus ring + translucent slide-in is preserved.
Tests: extend `e2e/a11y.spec.ts` with a Playwright check that the
desktop logo is clickable from its centre on the dashboard and the
URL stays at "/".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add an `<AppSettingsProvider>` that fetches `bugReportEnabled` once via
`/api/bugreport/status` and exposes it through `useAppSettings()`. The
sidebar consumes the flag and conditionally renders the Bug Report
entry — both in the collapsed icon mode and the expanded label mode.
The bottom-nav and topbar do not surface bugreport today, so no
changes were needed there.
The provider mounts inside `Providers` (above the toaster), so any
future flag-gated UI can reuse the same hook without re-fetching the
status payload. Defaults to `true` while the query is loading or the
user is unauthenticated so the entry doesn't flicker out under the
cursor on slow connections.
Tests: SSR-render `<SidebarNav>` with both flag states and assert the
`/bugreport` href appears / disappears accordingly.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pivot STATE.md + ROADMAP.md from the completed v1.4.14 marathon (live in
prod at /api/version=1.4.14, image digest 0ced46004a54) to the v1.4.15
scaffold: phases A1-A5 (quick fixes), B1-B6 (bigger features), C1-C5
(hardening including AI/Codex hallucination-resistance and Coolify
auto-deploy), D (multi-agent QA), E (release).
v1.4.14 entries archived in-place; docs/audit/v1414-summary.md remains
the canonical release record. Phase-0 report at
.planning/phase-0-report.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Marc-Brief (DE) at the top with what shipped, what's fixed, perf
deltas; below it the commit-by-commit landing report, deferred items,
CI/prod state, Codex-OAuth resolution, and the v1.4.15 backlog seed.
STATE.md flipped to finished.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- typecheck/lint/test/format clean (modulo the chart test file
blocked by Marc constraint — pre-existing)
- integration suite green after the test-isolation fix in the
preceding commit (10/10 consecutive runs)
- build + e2e blocked by the documented Node-25 Reflect.get bug
on this workstation; CI Docker (Node 22) is canonical and
Phase-3 already records 41/41 e2e specs green there
- format sweep on .planning/*.md + e2e/measurement-flow.spec.ts
to clear the warnings Phase 6 reconcile listed as
pre-existing-out-of-scope
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The integration suite runs with `vitest isolate: false` (one worker,
one container — migration cost demands it). Each test file declared
its own `vi.mock("next/headers", ...)` whose factory closed over a
file-local `cookieJar` Map. Vitest resolves a mock factory ONCE per
worker — whichever file loaded first won the registration, and the
other files' Map writes silently disappeared.
That made `admin-data-wipe.test.ts` flake intermittently (~12.5% per
empirical 8-run sample) when `idempotency-replay.test.ts` happened to
load first: its `vi.mock("@/lib/auth/session", ...)` then leaked,
returning a stub user without an admin role, so `requireAdmin()`
returned 401 and the wipe assertion failed with `expected 401 to be
200`.
Fix: extract cookieJar + headerJar into module-singleton Maps in
`tests/integration/mock-next-headers.ts`. All three integration files
import them and resolve them via dynamic-import inside the
`vi.mock("next/headers", ...)` factory body, so the factory always
sees the same backing Maps regardless of which file's registration
won the race. The `vi.mock("@/lib/auth/session", ...)` stub in
idempotency-replay was also removed; the test now seeds a real
Session row + cookie like the other two files.
After: 10/10 consecutive `pnpm test:integration` runs green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The v1.5.0 release was redirected to v1.4.14 because it is bugfix-heavy
with only additive features, so semver-patch is the correct increment.
Rename the v15-* audit notes and the v15-1 backlog accordingly, and log
the docs/landing rebrand (commits 06bc616 docs, 92c6588 landing) under
Phase 9 in STATE.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pre-phase-4b these sections were cards on the long monolithic /admin
scroll, where the expand-to-reveal kept 100+ table rows from
dominating the overview. Post-phase-4b they each own a dedicated
route reachable via the sidebar / mobile strip — the user has already
opted into the topic by navigating there. Landing on a card with only
a header and an "Expand" button reads as "is this page broken?".
Default `expanded = true`; the toggle stays as an escape hatch.
Closes H3 from .planning/phase-6-design-findings.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The four per-row action buttons on `/admin/users` (role toggle, edit,
reset password, force-logout) and the filter pills on both routes
were sized `h-7` (28px) — well below WCAG 2.5.5 AA's 44×44px target
that the rest of the app already honours (bottom-nav, dashboard
quick-add). Mobile users on a 360px viewport had four 28×28px buttons
crammed into the right edge of every user row.
Adds `min-h-11 min-w-11` to all six button sites. The visual icon-only
look is preserved — only the hit-area expands.
Closes H2 from .planning/phase-6-design-findings.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 4b's `<SectionFrame>` already paints the canonical h1 + subtitle
for each `/admin/<slug>` route, but every section card kept its v1.4
in-card h2 with the same (or near-same) text. The result was the
section title announcing twice — once as h1, then immediately again as
h2 — both visually redundant and a screen-reader landmark anti-pattern.
Demoting the in-card label to a `<div>` preserves the visual look
(same Tailwind classes) while removing the duplicate heading. The
four sub-cards inside `IntegrationsGroupSection` (umami, glitchtip,
web-push, bug-report) keep their h2 because they live under the
"Integrations" h1 and genuinely deserve a heading level — that is the
non-duplicating case.
Closes H1 from .planning/phase-6-design-findings.md.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
