0.8.1

What's Changed

[0.8.1] - 2026-09-04

A correctness pass over notifications and run artifacts: every webhook
payload now identifies the task it came from, notifications report the real
trigger, and the declared-but-dead capture_output setting is gone. Dark
themes get a precision pass of their own: CRT Dark replaces Gradient Dark,
and both dark themes now paint every page consistently.

Changed

• Gradient Dark is replaced by CRT Dark: an accent-phosphor dark theme
  with scanlines and an accent-tinted ambient wash (pick the green accent
  for a classic terminal). Users who had Gradient Dark are migrated
  automatically. CRT Dark's scanlines are drawn as faint light lines so the
  texture is actually visible on a near-black screen.
• Khaki Dark joins the dark roster — the warm brown twin of the light
  Khaki default — and High Contrast is now available in light mode too
  (near-black text on pure white with strong dark borders, mirroring the
  dark variant). Variants survive mode switches via the quick-toggle.
• Every surface, border, and text color across the app now derives from the
  theme tokens instead of hardcoded grays — previously only the Dashboard
  did. High Contrast (light or dark) and CRT Dark now look identical in
  quality on every page and dialog: true-black panels, light borders,
  properly bright text. Light themes are visually unchanged.

Fixed

• Webhook notifications name the task and its outcome. Payloads read like
  the desktop toast — "deploy — Failed • Trigger: schedule • Duration:
  5.1s" — instead of a bare "Trigger: … • Duration: …" that gave no way to
  tell which task ran.
• Scheduled runs no longer notify as "Trigger: manual". The completion
  callback now receives the actual trigger (manual, schedule, cli, system)
  through the executor's group result.
• Missed-run reconciliation and the NEXT RUN display evaluate cron specs in
  the daemon's local timezone. Stored windows are UTC while the running
  engine fires local ticks, so on a UTC+6 machine a daily 9:00 schedule that
  fired while the PC was off reported "0 caught up" at boot and silently
  dropped the missed run.
• Webhook format "pumble" validates again. Pumble incoming webhooks accept
  Slack-style payloads, so the editor now offers a single "Slack / Pumble"
  option that saves the canonical "slack" format; existing YAML with
  "pumble" loads as-is and re-exports as "slack".
• Per-run artifacts (stdout.log, stderr.log, run.json) land under the
  global runs folder (<data_dir>/runs, configurable via
  run_artifacts_dir) unless a task sets output_dir. They previously went
  to the task's working directory, leaving the configured runs root — and
  the "run artifacts" part of backups — empty.

Removed

• The capture_output task field. It had no behavioral effect (output is
  always capped and stored on the run row). Existing YAML files keep
  parsing; the field is dropped the next time the task is saved.

[0.8.0] - 2026-09-04

Backup & restore, plus a proper home for a growing vault. Every archive is an
AES-256-encrypted zip that captures the complete Heka state — a consistent
SQLite snapshot, the vault key, the canonical task YAML files, and the user
config — so a restore brings back exactly what was there, secrets included.

Added

• Automatic backups on a configurable schedule — every N hours (up to a
  year), daily, weekly, or monthly at a local time — with missed-run
  catch-up: if Heka was closed at the scheduled time, the missed backup
  runs once at startup.
• Local destination (custom folder + keep-latest retention, default inside
  the data directory) and an S3-compatible remote destination — AWS S3,
  Cloudflare R2, Backblaze B2, MinIO — with its own keep-latest pruning and
  a one-click "Test connection" that exercises the bucket end to end.
• S3 credentials live in the secrets vault (BACKUP_S3_ACCESS_KEY_ID,
  BACKUP_S3_SECRET_ACCESS_KEY), same write-only discipline as every other
  secret.
• Archive encryption with a passphrase stored in the vault
  (BACKUP_PASSPHRASE) so scheduled backups run unattended — with a loud
  warning that losing it means losing the backups, since restore on a new
  machine asks for it.
• Manual "Back up now" with live status, a job history view (trigger,
  status, size, per-destination outcomes), and desktop notifications when a
  scheduled backup fails.
• A guided restore flow: pick an archive, verify it with the passphrase,
  preview exactly what it contains (created when, which Heka version, task
  counts), choose optional parts (config.yaml, run artifacts), and let the
  app stop the daemon, restore, and prompt to start it again. A safety
  backup of the previous state is written automatically before anything is
  overwritten; archives from newer Heka versions are refused, and checksums
  guard against truncated or tampered archives.
• A dedicated Backup section in Settings with all of the above, and
  heka backup create|status|history|test plus heka restore in the CLI.
• A dedicated secrets manager page (/secrets, linked from Settings →
  Secrets) built for large vaults: search, sort, "unused only" filter,
  per-key usage tracking (which tasks reference ${KEY}), bulk delete with
  confirmation, pagination, and per-key value replacement. Keys stay the
  only thing ever displayed — values remain write-only.

Changed

• Backup catch-up is now part of the missed-run reconcile pipeline: a missed
  auto-backup runs once on daemon start, on wake from sleep, on the periodic
  reconcile cadence, and from the "Reconcile now" button — not only via the
  backup loop's own tick. A reconciled backup is labeled "catch-up" in the
  job history, and the missed window is recorded in the system log
  (Logs → System), alongside new events for manual starts and schedule
  arming.
• The backup time-of-day picker is a themed segmented field (HeroUI
  TimeField) instead of the native browser time dropdown, which overflowed
  the panel.

Fixed

• The Schedules page no longer shows a stale NEXT RUN after a missed run
  was caught up. Reconcile closed the miss window without re-deriving
  next_run_at, and the startup path read a cron entry that only carries a
  value once the engine starts — so a daily 9:00 schedule kept showing
  yesterday's 9:00 as next. Both paths now compute the next occurrence
  directly from the cron spec.
• S3 "Test connection" now saves pending edits first (endpoint, bucket,
  keys, Use HTTPS), so it tests exactly what is on screen. Enabling the S3
  destination also defaults to HTTPS — plain HTTP is 301-redirected by
  Cloudflare and friends, which previously surfaced as a wall of HTML; the
  redirect now produces a clear, actionable error instead.

Full Changelog: 0.7.7...0.8.1