perf(ci): cut about a minute from every release

[t3dotgg]

Releases take a median 13m37s. Two pieces of that are pure waste: the CLI publish job builds the web client twice, and two startup jobs wait for version resolution they never use. This PR removes both, worth about a minute per release. It also fixes two scheduling problems found in the same audit.

• publish_cli built the web client twice. The t3 build task already depends on @t3tools/web#build, so the explicit web build step repeated ~35s of work on the serial publish path. Verified locally: vp run --filter t3 build emits exactly one web build and still produces dist/bin.mjs, dist/service-launcher.mjs, and dist/client/index.html.
• relay_public_config and build_wsl_node_pty now run alongside preflight. Both consume only the commit SHA. They previously waited ~30s for preflight to resolve the version, which they never read. Desktop builds still hard-gate on preflight and relay config succeeding.
• Nightly cron moves to minute 7. Scheduled runs started a median 13m late across 56 sampled runs. GitHub documents the worst scheduling delays at the top of the hour.
• Concurrency groups per channel. Overlapping nightlies double-built the same commit on August 23. Nightlies now queue behind each other with cancel-in-progress: false, so no running publisher is ever canceled. Stable tag releases get their own group and are never blocked by a nightly.

Timing data comes from an audit of 60 release runs. The larger follow-ups from that audit (prebuilt hosted-web deploy, early CLI build, Windows cache benchmark) are separate work.

Changes by Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Low Risk
Changes are limited to GitHub Actions release orchestration; artifact contents should be unchanged aside from avoiding a redundant web build.

Overview
This PR trims release workflow wall time (~1 minute per run) and hardens nightly scheduling.

Parallelism: relay_public_config and build_wsl_node_pty no longer wait on preflight—they only need the commit SHA, so they start in parallel with version resolution (same gating as preflight on scheduled runs when check_changes finds new commits). Both jobs check out github.sha instead of a preflight ref.

Duplicate work: publish_cli drops the explicit @t3tools/web build step; vp run --filter t3 build already pulls in the web client via the existing task dependency.

Scheduling / safety: The nightly cron moves from minute 0 to minute 7 every 3 hours to reduce GitHub’s top-of-hour delay. A concurrency block serializes nightly runs (cancel-in-progress: false, queue: max) while stable tag releases use a separate group so they are not blocked or dropped when nightlies queue.

^{Reviewed by Cursor Bugbot for commit ebe3ebf. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Run relay_public_config and build_wsl_node_pty in parallel with preflight in release workflow

• Changes relay_public_config and build_wsl_node_pty to depend on check_changes instead of preflight, so they run in parallel with preflight rather than waiting for it
• Both jobs are skipped on scheduled runs when has_changes is not true, avoiding redundant builds
• Checkout now uses github.sha directly instead of needs.preflight.outputs.ref
• Removes the explicit "Build web package" step since the CLI build already produces it
• Adds concurrency grouping by channel (nightly vs stable) with cancel-in-progress: false, and shifts the cron schedule from minute 0 to minute 7 to avoid GitHub peak load
• Risk: jobs no longer wait for preflight to complete, so a preflight failure will not prevent these jobs from starting; verify that relay_public_config and build_wsl_node_pty do not rely on any output from preflight beyond ref

^{Macroscope summarized ebe3ebf.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 17aef292-c2bd-4228-b780-55f0ee66b8ee

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.3 KiB	13.3 KiB	+24 B (+0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−2 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+26 B (+0.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	11	+1 (+10.0%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	−20 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+7 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−27 B (−0.4%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.4 KiB	56.4 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	11	10	−1 (−9.1%)	21	✅

Baseline: b0a0281 · PR result: ebe3ebf · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — The PR changes release scheduling, concurrency, queueing, and job dependency behavior for workflows that publish production artifacts. The redundant web build removal is localized, but the release-orchestration changes have material operational impact and should receive human review.

^{You can add or adjust custom eligibility rules. Learn more.}

Dismissing prior approval to re-evaluate ebe3ebf

[t3dotgg]

Note

🤖 Claude Fable 5 responding on behalf of Theo

Merging over this verdict. It asks for human review of the release-orchestration changes, and that review happened: Theo and a second reviewing agent audited the workflow changes, the reviewer flagged the pending-stable cancellation edge case, and ebe3ebf fixed it with queue: max per Theo's explicit call. Macroscope's own correctness check passes on this commit, and no reviewer found a defect. The remaining risk is runtime behavior GitHub only exercises on real triggers, which the next scheduled nightly will confirm.