Skip to content

wheeljack v0.1.14

Latest

Choose a tag to compare

@github-actions github-actions released this 05 Sep 10:56
· 2 commits to main since this release
837e3ad

What’s new

  • Complete backup and restore. Save wheeljack state and referenced image attachments in a portable, checksummed folder. Preview a backup before restoring it on the next launch; wheeljack saves a complete recovery backup of the previous state first. Attachment paths relocate correctly, including macOS path aliases. Project files and provider credentials remain separate. Keep the whole backup folder together.
  • More reliable agent recovery. Pending prompts follow resumed sessions, idle transcript tails are saved, and accepted prompts are not resent when local persistence needs to retry. Editing a queued prompt preserves your separate message draft and attachments.
  • Clearer workspace controls. Model controls reflect each pane’s effective configuration. Home actions identify their project, and counts include task worktrees consistently. PRD/TDD actions now say Add starter tasks and add generic checklist cards without duplicates.
  • Better previews and lifecycle handling. Local previews wait for HTTP readiness and support explicit reload. Lifecycle timeouts continue to cover descendants holding output pipes open, and multiple agent-control lines are handled independently.
  • Less background work and smaller JavaScript. Empty queue workers retire, retention runs while the app stays open, and Settings/Ops load on demand. Initial JavaScript is 44.7% smaller in the measured build.

Compatibility

The delivery metadata upgrade preserves database compatibility with v0.1.13, including rollback after an unsuccessful update. Completed/canceled prompt bodies expire after seven days; unresolved work and minimal delivery identities remain preserved.

Validation

Validation includes 366 frontend tests; 302 core tests on Windows and 292 on macOS; 17 host tests on each platform; native app, update, and rollback checks on both platforms; a live Codex queue/resume check; and packaged Windows backup and previous-release upgrade scenarios. In the measured Windows six-session benchmark, input and frame targets pass; memory remains about 665 MiB against a 540 MiB target. Performance comparisons use the recorded historical baseline, not a same-day paired benchmark.

Downloads and provenance

macOS universal app and DMG: signed with Developer ID and notarized by Apple. Windows portable executable: unsigned. All downloads have SHA-256 sidecars; SHA256SUMS.txt covers all three packages.

Source: 837e3ad. Changes: PR #20.