[miguel-heygen]

What changed

Release v0.8.212: the version bump for all 14 packages and the 5 plugin manifests, the release notes and the changelog entry.

• A model download that stalls partway resumes instead of failing (fix(cli): resume a model download that stalls partway instead of failing #6649).
• Failures with --json answer in one shape, and --json is accepted only where a command supports it (fix(cli): every --json failure answers with one envelope, and --json only where supported #6622).
• check prints a progress line as each pass starts, and its WebGL page check no longer keeps a list of 2d canvases (fix(cli): check prints a progress line as each pass after overlap begins #6646, fix(cli): a WebGL canvas stays readable on a check page, so its motion is not read as dead #6650).
• Also in the range: the render route passes a MOV ProRes profile (fix(studio-server): the render route passes a MOV ProRes profile to the render #6644), a Windows browser that will not start says where and why (fix(engine): a Windows browser that will not start says where and why #6647), help [cmd] with closest-command suggestions (fix(cli): add help [cmd], suggest the closest command or flag, and quiet the nags #6621), live inspector queries across compositions (fix(inspector): live queries see every composition, wherever the playhead is #6651), a thumbnail reader that forgets failures after a minute (fix(studio): a thumbnail reader forgets a start's failures after a minute #6642), and two CI/test changes.

This PR also corrects the v0.8.211 notes (releases/v0.8.211.md and its changelog entry). The published 0.8.211 contains 17 changes its notes did not list, because its release PR merged through the queue onto a main that had moved: renders encode while capturing, cloud renders stitch in parallel, more capture workers on Macs with a GPU, one remote download per render, the audio mix running alongside browser start-up, drawElement films keeping their workers, Windows renders opening no console windows, Windows undo and redo on a held file, a named reason for a choppy preview, and more. They are now listed, in plain words.

What I measured

• bun run release:prepare 0.8.212 --from v0.8.211 on a full clone drafted the notes from the 11 commits since v0.8.211; I wrote the summary and left the generated lines as they are. It then set 14 packages and 5 plugin manifests to 0.8.212, with --no-tag because the workflow tags.
• The commit hooks (format, commitlint, tracked artifacts) pass.
• The v0.8.211 correction was checked against git log v0.8.210..v0.8.211 (the real tag range), where the first notes were checked against the range the PR opened with.

What I did NOT exercise

• The publish: the workflow tags and publishes after the merge, so npm is unchecked until then. The 0.8.211 publish ended red on one package's signing-log entry (a transient 409) and needed one rerun of the failed job.

Proposed, not built: notes generated from the real range

The 0.8.211 notes were written against the range at PR-open time and the queue then merged the release commit onto a later main. Two places could own the fix, and I would pick the second:

1. In release:prepare, regenerate at merge time: not possible there, because the script runs before the queue.
2. A check on the release PR and again on the merge-group run: list the commits from the last tag to the base of the merge group, and fail if any change in that range is missing from releases/vX.md. It sees the range as the queue will tag it, and it fails the PR in the queue instead of after the publish.

Want it built? It is a small script plus one CI step, and it needs your call on whether a miss should fail the queue.

No visible change

A version bump, release notes and a changelog entry; nothing here changes what the studio or the player shows.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
hyperframes	🟢 Ready	View Preview	Oct 28, 2026, 12:05 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

[heygengenesis]

Verdict: APPROVE · 0 blocker · 0 important · 1 nit · f644e31

• nit: the 17 v0.8.211 bullets added in both releases/v0.8.211.md and the mdx entry have no commit or PR links. Every other generated bullet links both (e.g. #6631, #6643, #6616).

Scope

The head's parent is origin/main (85efaec), so the notes match the range the queue will tag. All 11 commits in v0.8.211..main are listed in releases/v0.8.212.md and the mdx <Update>. The v0.8.211 correction covers all 17 commits in the real v0.8.210..v0.8.211 tag range that were missing: 2 features, 6 fixes, 9 perf. All 14 non-private packages and the 5 plugin manifests are bumped 0.8.211 → 0.8.212. CI, regression, Windows render verification and every other workflow on this head are green. I'm leaving the proposed range check in the body for a human to decide.

[heygengenesis]

Verdict: APPROVE · 0 resolved, 0 open, 0 new · f644e31 → c748722

Rebuilt on main (2d3033c). The 13 PRs in v0.8.211..main each appear once in releases/v0.8.212.md and once in the mdx entry, including the new #6639 and #6656. All 14 packages and 5 manifests are at 0.8.212, and every workflow on this head is green. The earlier nit still stands: the corrected v0.8.211 bullets have no links.