Repository navigation
chore(sync): merge upstream v2.3.0 and prepare 2.3.0-shuv.1 - #489
Conversation
…es (anomalyco#63503) Co-authored-by: opencode-agent[bot] <219766164+opencode-agent[bot]@users.noreply.github.com>
…cs previews (anomalyco#63560) Co-authored-by: Adam <2363879+adamdotdevin@users.noreply.github.com>
Co-authored-by: opencode-agent[bot] <219766164+opencode-agent[bot]@users.noreply.github.com>
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
Independent review: findings addressedRequested Cursor/Claude Opus 5.5, effort high, with approved pool failover. Review ran read-only after this PR opened (resolved seat: Cursor-8). The reviewer inspected the complete local merge/remerge diff and independently reproduced the recovery behavior against the old and merged trees. Two high-priority findings:
Reviewer verified clean: normal merge ancestry, all 12 conflict resolutions outside recovery, terminal mentions/paste undo, fork Evaluation exports, E2E fixture adaptations, managed-policy waits, all three hidden-provider layers, package/bin/version identity, verbatim upstream notes/preserved fork history, workflow exclusions and whitespace. Both findings are addressed. This is an independent review result with implementation follow-up, not a GitHub platform approval or a claim that the reviewer re-reviewed the final fix. Final full-suite/build/check results are maintained in the PR body and |
Summary
Normal two-parent merge of upstream v2.3.0, following #484 and #487, with 2.3.0-shuv.1 release preparation in this PR. Targets only
shuv1337/shuvcode:integration-v2. No default-branch push, rebase, force-push, tag, release, publish, deployment or live-service change. Stop before merge.Ancestry
ac4077949d227a1ff244a6f9f2b398ccf14732ef, including chore(sync): merge upstream dev through 101776024c #487's final fixes and chore(release): 2.2.7-shuv.2 #488.101776024cc69bafb3f9dc90027ab4d1eab40a2a.a6b15d33daf6ab7d35569f6d711fdb630b9f04e7(v2.3.0), November 13, 2026 at 12:55 AM PST; 70 commits past that pin. Parent73abe63ac23224d88e739e7028882e547ef65a79.93364ed1a7a1ad59a55bc75ae5645f85ac2e858b. Newer upstream dev commits are outside this release sync.a27a4dc3aa8a1d14eefa94c5ae10b24e0f0a995c, open and unmerged.Durable audit and complete branch dispositions.
Conflict resolutions
Twelve paths conflicted:
.github/workflows/release.ymlAGENTS.md,specs/v2/session.mdCHANGELOG.md,packages/cli/package.json,bun.lockpackages/ai/src/experimental.tspackages/app/e2e/regression/timeline-{geometry,stream}.spec.tspackages/core/src/session/execution/restart.tspackages/core/test/session-execution.test.tspackages/tui/src/component/prompt/index.tsxManaged policy and fork preservation
Keep #484's upstream behavior:
model.list,model.default,provider.list,generate.*,voice.*,websearch.querywait up to 5 seconds, then return 503 if activation is incomplete. Other catalog reads remain non-waiting. Server policy code is unchanged from the fork base; no extra waits added.Preserve Shuvcode XDG/ports, npm-only updates/publication guard, bundled Claude/Antigravity/Codex/UI plugins, all three Claude Agent SDK hiding layers, tool-activity leases, rich failures, inactivity interruption and headless configuration diagnostics. Claude Agent SDK fixes stay on disk only.
Validation
Bun 1.4.4. Runtime checks use unprivileged namespaces, fresh HOME/XDG/temp roots, an empty environment allowlist and hidden real configuration/credentials/service state. Tests use private loopback networks; build downloads allow network. Synthetic /dev and private /proc only; no privileged/systemd container or host /sys, /run/udev or /run/dbus mount.
bun run checka27a4dc3aaFinal product source:
c1d4594b55. The initial full run caught a genuine recovery interaction: the direct-claim sweep could fail an ambiguous child before prepared-parent recovery, allowing the parent to resume. The follow-up carries ambiguity into prepared recovery and stops the parent too. Existing regression assertions pass ten repeats each and the full gate is green; no tests/timeouts/skips were weakened. The initial failing run reproduced the known Turbo failure-path SIGSEGV (pc_4bef4e2422a7); the green run exits normally.The initial hygiene finding was a literal local home path in the audit, changed to
$HOMEwithout weakening the allowlist. Two batched Node-smoke invocations exited 141 with no log; unchanged standalone reruns pass, including the final build (pc_0cf7e3fd2890). The first post-review full run under concurrent build/typecheck load timed out in an unchanged client service-eviction test; ten standalone repeats pass and the final full gate passes without concurrent builds (pc_45f7a80dd7f4). No timeout/retry/skip changes were made.No browser E2E/component suite, cross-platform execution or all-target build claimed. ## Independent review
Claude Opus 5.5, high effort, completed the requested read-only review via approved Cursor pool failover. Two high-priority findings, both addressed:
1ddafa6b72, independently verified by the reviewer.c1d4594b55, excluding the precomputed prepared-child set from the owner loop. New regression failed before the fix and passes afterward, including a second sweep; all four cases pass 40/40 repeats.All other reviewed fork-preservation/conflict areas were clean. Detailed review and resolution. No GitHub platform approval or independent re-review of the final fix is claimed.
Feature flags
No new fork flag for sync/release metadata. Existing upstream experimental flags remain; stable video generation follows upstream graduation.
Reviewing this large merge
Over 300 files:
git fetch origin pull/489/headthen review the complete localgit diff ac4077949d...FETCH_HEAD; do not rely on GitHub's capped diff endpoint. Inspectgit show --remerge-diff 93364ed1a7for conflict resolutions.