Releasesv0.29.0
v0.29.0 — Code screen: plan mode, verify output, per-run model
This is the release note as published, not a rewrite of it. Release notes are published in the language they were written in.
Three honest-by-construction upgrades to the desktop Code screen — converging on what Claude Code and Antigravity surface, without faking any signal the backend can't observe.
What's new
- Plan mode (preview + gate). A new
POST /api/planruns only the planner — no edits, no tools — and returns the concrete steps. Review and edit them, then Run with this plan (used verbatim, re-planning skipped) or Run without. Because the preview makes zero file changes, approving it is the pre-edit gate. The steps double as the run's live task list. - Real verify output. Each attempt now shows the verifier's actual captured stdout/stderr in the receipt (only when non-empty) — the real command output the loop already recorded, not just pass/fail.
- Per-run model & mode. Pick a model and
single | fuse | cascadeper run, right in the Code panel. The default (no flags) is byte-identical to before.
Everything stays optional and the terminal CLI stays sovereign. Native installers (.exe/.dmg/.AppImage/.deb) are attached by CI a few minutes after publish; also pip install 'chimera-agent[desktop]'.
Full changelog: https://github.com/brcampidelli/chimera-agent/blob/main/CHANGELOG.md