v0.3.4.

Added

• One-Click Panel Self-Update: operators can upgrade the panel from the dashboard. New GET /v1/system/update-status and POST /v1/system/update-start endpoints run this install's own install.sh --version <tag> for an operator-pinned exact release tag; on systemd hosts the updater launches through systemd-run into a transient cgroup so it survives stopping the unit it belongs to, state lives in marker files under <dataDir>/self-update/, and the updater's environment is deliberately narrow (no JWT/DB secrets reachable via systemctl show). A layout banner plus About-card button share the new usePanelUpdate poller, whose phase survives panel restarts via localStorage and reports installer output tails on failure.
• Volume Backup Labels: manual volume snapshots accept an operator label (≤40 chars, default manual) and scheduled runs are labeled schedule-YYYY-MM-DD; migration 0037 persists backups.label, serialization exposes scope/volumeName too, and the Backups page names every row instead of showing bare timestamps.
• Webhook URLs Honor The Panel Domain: auto-deploy payload URLs are now built from the Settings → Security "panel domain" (panel_domain), falling back to NINEDEPLOY_DOMAIN then NINEDEPLOY_PUBLIC_URL; scheme mirrors the Traefik router (https only when an ACME email exists). The environment tab warns amber when stored webhook URLs point at localhost so git providers cannot silently fail delivery.
• Honest Deployment History: finalizing a successful deploy demotes older running rows to a new superseded status, and boot-time reconcileDeploymentHistory keeps at most the newest running deployment per actually-running service — past deploys stop displaying "Running" forever.
• Cron-Preset Jobs Editor & Raw Env Editing: the scheduled-jobs card is rebuilt around presets (time/weekday/monthday pickers, custom-cron input with validation) backed by a new frontend lib/cron.ts (parseCron, nextCronRun, describeCron); env vars gain a table ⇄ raw .env bulk-edit mode accepting comments, blank lines and quotes, saving added/updated/removed diffs in parallel.

Changed

• Detaching A Volume No Longer Requires Stopping The Service: detach now queues a blue-green recreate without the mount, matching attach/update behavior since Docker cannot hot-swap -v.
• Alert Rules Expose lastEvaluatedAt so the monitoring page can distinguish "never evaluated yet" from a lapsed collector.

Docs

• README, docs/ and the website were re-verified against the code: MCP tool count (35), nine managed-database engines, 88 hub templates with 16 runtime-certified, kernel event/hook names, the supported ACME DNS-provider list, Docker-mode upgrade instructions, OIDC configuration path (admin UI, not env vars), and coverage floors stated per package instead of a blanket 100% claim.
• The README system-architecture diagram now matches the engine: three workload types (containers, PM2 processes, Compose stacks), nine database engines, Traefik routing published apps in addition to the panel, the HMAC webhook ingress path, Cloudflare Tunnel as a parallel ingress for NAT-restricted nodes, and the new panel self-update machinery.

Fixed

• Every CI Job Died Before Running A Single Test: pnpm/action-setup was pinned to 11.22.0 in the workflows while packageManager declares pnpm@11.23.0 — action-setup refuses the conflict with ERR_PNPM_BAD_PM_VERSION, which is why "Typecheck · Lint · Build · Test" failed in seconds alongside everything else. The workflow pin is removed; the version derives from packageManager (the same pattern website.yml already used successfully).
• The Release Workflow Would Have Failed The Same Way On The First v0.3.4 Tag: release.yml carried the identical pinned-setup block and is fixed identically.
• Node 26 Images No Longer Bundle Corepack: the Dockerfile's RUN corepack enable (both build and runtime stages) died with /bin/sh: corepack: not found, breaking the CI "Docker image build" job and any tagged release image. Both stages now install a pinned pnpm via npm (ARG PNPM_VERSION), kept in sync with packageManager.