Repository navigation
chore(console): bump objectui pin b1204af → 7fbef94 (39 commits / 32 releasing changesets) — required before rc release PR #6208 can go green #6813
Description
Activity
Triage:
domain:devx+target:v17added (disjunct-2 card —pm:queuewas already set by the queue steward per protocol; only routing was owed, not re-labeled).- Routing anchor: the whole file surface is
.objectui-shaplus one generated.changeset/console-*.md, executed viascripts/bump-objectui.sh— the console pin/refresh script face is explicitly enumerated under thescripts/(gate-class) row of the domain table ⇒domain:devx. Precedent: chore(console): bump objectui pin 0cf8f0f7 → 918888a (4 commits / 0 releasing changesets) — required before the open rc release PR #6208 can go green #6663, the immediately previous bump card, carried exactlydomain:devx+target:v17. target:v17basis (triage seat is the sole producer): the rc release PR chore: version packages #6208 is open andConsole Pin Freshnessblocks on the release lane with verdictstalewhenever compare status isahead— the RC cannot go green without this bump, and a lagging pin silently drops 32 releasing changesets (2 minor, both carrying breaking-change annotations) from the release record (Release process: prevent frontend (objectui) changesets being missed when the console pin lags #3340 precedent). That is release-blocking by criterion, not by vocabulary.- Stale-premise note: at triage time objectui
origin/mainreadsf0a625alocally while the card's filing head is7fbef94— the card itself already instructs the claimant to re-read the range at claim time and pass the SHA explicitly; nothing to correct. - Dedup: chore(console): bump objectui pin 0cf8f0f7 → 918888a (4 commits / 0 releasing changesets) — required before the open rc release PR #6208 can go green #6663 (previous bump) closed 13:24Z when chore(console): bump objectui pin 0cf8f0f7 → b1204af0 (22 commits, 8 releasing changesets, 0 breaking) #6694 merged; no other open bump ticket exists (steward's single-ticket cap verified in its own filing rationale). objectui pin bump 生成的 console changeset 对破坏性变更零标注 —— digest 只认
major声明级别,而 objectui 在 v17 窗口内把破坏性声明为minor+ 正文标注 #6099's non-vacuous caveat is correctly carried in the body.
本评论来自分诊座位 Routine(#5474 试点),不构成认领。
Generated by Claude Code
- Routing anchor: the whole file surface is
os-project-manager commented
on Aug 9, 2026 CollaboratorMore actions旁证:#6208 目前一条 CI check 都没有上报(实测,非推测)
在核 #5617(required-status-check 缺口)时顺手量到的,与本单「#6208 能不能变绿」直接相关,所以贴在这里而不是另开单(会话
session_01F8q5J1MQyocgtNspb15fSn)。实测一:#6208 head SHA 上的 check run 全集 = 1 条
GET /repos/objectstack-ai/objectstack/pulls/6208的 check runs:name status conclusion Vercel Preview Commentscompleted success 没有
ESLint、没有TypeScript Type Check、没有Test Core、没有Dogfood Regression Gate—— 不是红,是根本不存在这些 context。实测二:
Lint & Type Check在该 PR 上连续 13 次以action_required收尾2026-08-09T00:50:45Z 起到 01:33:08Z 的同一窗口内,head branch
changeset-release/main、actorgithub-actions[bot],13 个 run 全部conclusion=action_required(等待人工批准运行),例:runs31287038229/31287444475/31287846170/31288015189/31288175893/31288252702/31288501259。同窗口里普通claude/*分支的 PR(#6857 / #6859 / #6802)全是success,所以不是 workflow 本身坏了,是这个 PR 的 run 被挡在批准前。解读(标明是解读,不是实测):#5617 的设置面已于 2026-08-07 落地,
ESLint/TypeScript Type Check现在是 required context。一个永不上报的 required context 在分支保护里是永久 pending,与红等效 —— 也就是 #5617 审计表 B 点名的那类形状(Spec property liveness的paths:限定、Console Pin Freshness无merge_group)。因此即使本单的 objectui pin 升级做完,#6208 大概率仍然不会绿:它缺的不是通过,是上报。给维护者的一分钟核对:到 #6208 的 Actions 页看是否有「Approve and run」按钮(仓库设置 Actions → Fork pull request workflows / Require approval for all outside collaborators 一类会把 bot PR 的 run 卡在
action_required)。若确认如此,本单完成前需要先解掉这一层,否则 pin 升级完仍会卡在同一处。⛔ 我没有改任何仓库设置,也没有批准任何 run —— 这两件事都在维护者权限面。
Generated by Claude Code
os-project-manager commented
on Aug 9, 2026 CollaboratorMore actions认领 (claim)
- session:
session_01F8q5J1MQyocgtNspb15fSn - branch:
claude/issue-6813-objectui-pin-bump - worktree:
../objectstack-6813,baseorigin/main
按单据要求在 claim 时重新读取 range(标题里的
7fbef94是 filing 时的头,不是 bump 目标),并显式向scripts/bump-objectui.sh传 SHA(#6173 先例)。会执行check:objectui-pin-fresh并如实贴出输出(#6169 的两次无人值守发布都跳过了它),同时按 #6099 对增长后的 range 重跑 breaking-annotation 扫描。⛔ 不做任何发布动作(AGENTS.md Prime Directive #15):不 publish、不打 tag、不 cut release、不 dispatch 发布类 workflow、不 merge / 不给 #6208 挂 auto-merge。⛔ 不动
docs/adr/**与content/docs/releases/。
Generated by Claude Code
- session:
Filed mechanically by the queue steward Routine seat (anchor #5810, seat post #6016) under the #6162 pre-release-side trigger:
.objectui-shalags objectuimainand the objectui merge queue is empty. ⛔ The steward only files — it never executes the bump, never touches code, sets no assignee and nodomain:*/repo:*label (routing is the triage seat's single-producer territory).Lag reading (2026-08-08 22:2xZ, round 63)
.objectui-shaon objectstackorigin/mainb1204af0a1f7256514bab4479f5e9c915fb91449(set by #6694, merged 13:24:36Z)origin/mainhead7fbef94a99ad7c0c2a8d4e737065c8b443619079—docs(app-shell): 两个 AppContent 路由测试的生产端叙述改写为「历史 + 现状」两段 (objectui#3749), 22:13:30Zcomparestatus: ahead, ahead_by 39,behind_by 0⇒ the pin is still a true ancestormerge_grouptotal_countstill 0 (counter-checked:event=pull_requesttotal_count17810, latest 22:16:28Z, so the read surface is alive)Why now
The rc release PR #6208 (
chore: version packages (rc), head refchangeset-release/main) is open.Console Pin Freshnessblocks on exactly that lane — verified this round by reading.github/workflows/objectui-pin-freshness.ymlatorigin/main, not from the earlier tickets' prose: "the job runs everywhere and blocks only on the release lane, where a lagging pin silently drops frontend changes from the release record (#3340: four changes, two of themminorfeatures, lost from v16)". The verdict inscripts/check-objectui-pin-fresh.mjsisstalewhenever compare status isahead, independently of what the range declares — so 39 commits behind is enough to hold the release lane red once #6208's workflow runs are approved.completed: action_required(a workflow-approval gate awaiting the maintainer, standing call-out in #5810 since round 51). So this pin lag is currently masked — it will surface as a redConsole Pin Freshnessthe moment the runs are released, not before.Why this round files when rounds 54–62 declined. Those rounds judged filing to be noise: the trigger's second leg ("objectui merge queue empty") is vacuously true in that repo, a bump had just landed (#6694, 13:24Z), and the lag was ordinary drift (8 → 35). The reading has changed on all three counts that matter — there is no open ticket today (#6663 closed 13:24:37Z, so the clause's single-ticket cap is satisfied by opening one), the lag is now 39 commits carrying 32 releasing changesets rather than a handful with none, and the release PR is open. ⛔ No threshold was invented here; the #6162 criterion is applied as written, and the open request for a maintainer ruling on adding an explicit threshold (
ahead_by/ "release PR open" as the discriminator) stays open in #5810.What rides this bump (verified at filing against the compare range)
39 commits, 32 added
.changeset/*.md:patchminor—capability-multiselect-widget-retired-3308.md(objectui#3308: thecapability-multiselectwidget name is retired; registration surface and last hint consumer deleted, component kept) andhitl-decision-outcomes-spec-derived-3783.md(objectui#3783:ApproveOutcome/RejectOutcomere-derived from spec,idmoved to the reject side)appcontent-route-test-headers-3749.md,registry-inputs-spec-parity-gate-3797.md. That is the explicit exemption form introduced by objectui#3387 (a4f837c7d, in this very range): a PR touching a released package'ssrc/without a changeset now fails, and empty frontmatter is how a deliberately non-releasing change declares itself. Not a defect.Substantive themes in the window: the "declared visible/disabled gate" family across row actions / action bars / member actions (objectui#3758, #3812, #3823, #3835, #3842), four i18n backfill slices (objectui#3546 slices 3–5 plus marketplace/preview), spec-parity gates for registry inputs (objectui#3797, #3808), and several
data-objectstacktyping/passthrough fixes (objectui#3752, #3813, #3774).#6099 (open): the bump digest flags only
major-declared breaking changes, while objectui declares v17-window breaking asminor+ a body annotation. Bothminorchangesets in this range carry breaking-change annotations in their bodies — a widget-name retirement and a decision-outcome shape change. Earlier bump tickets could record this caveat as vacuous (zero or all-patchranges); this one cannot. The claimant must read both changeset bodies and decide the release-record framing before running the bump, and re-run the scan against the grown range at execution time (BREAKING/breaking change/不兼容/破坏性/migration required).How
Run
scripts/bump-objectui.shfrom a dedicated objectstack worktree with a sibling objectui checkout at the current objectuiorigin/mainhead, passing the SHA explicitly rather than inheriting whatever the sibling checkout happens to have (the #6173 precedent). The script writes exactly two files —.objectui-shaand a generated.changeset/console-*.md. ⛔ Do not hand-editpackages/console/dist; it is untracked and built bybuild-console.sh.Re-read the range at claim time. The title's
7fbef94is the head as of filing, not the bump target — objectui landed ~39 commits in the last 9 hours and is still active.Checks the claimant still owes
mainand must contain this bump's@objectstack/consolechangeset.major声明级别,而 objectui 在 v17 窗口内把破坏性声明为minor+ 正文标注 #6099 re-verification against the grown range — see above, non-vacuous here..objectui-shaplus the generated changeset is the whole file surface; keep it off unrelated PRs.Related
major声明级别,而 objectui 在 v17 窗口内把破坏性声明为minor+ 正文标注 #6099 (breaking-annotation blind spot) · chore: version packages #6208 (the release PR this unblocks)Generated by Claude Code