Repository navigation
console: the record ⋯ overflow menu's destructive band honours the spec's variant: 'danger' (objectstack#24277 ruling E1) #12961
Description
Filed by the director seat (summon #43 continuation 5, session_01QJC7i981TSs9L1CGJK59Ax) from objectstack-ai/objectstack#24277, ruled E1 by the maintainer on 2026-10-18 (batch #365 item 1): keep the header fold and the ⋯ menu as they are — no new spec key, no grouping — and make the menu's existing destructive band honour the spec's own spelling. ⛔ Not a claim.
The defect (read from source at objectui a8560b228527; not yet measured in a browser — the first step below measures it)
packages/components/src/renderers/layout/containers.tsx(PageHeaderRenderer, the overflow bands of objectui#12529, around line 2075):isDestructiveAction = (a) => a?.variant === 'destructive' || a?.name === 'sys_delete'. The spec'sActionDefinition.variantenum isprimary | secondary | danger | ghost | link(@objectstack/specpackages/spec/src/ui/action.zod.ts, around line 1310) —destructiveis the Button primitive's word, not a legal action variant. Nothing on the header path mapsdangertodestructive(DeclaredActionsBar.tsx:409does, for its own buttons only).- Consequence, if measured true: an object's own
variant: 'danger'actions (cloud'ssys_environmentarchive and delete,sys-environment.object.ts:1162and:1221;sys_user'sdelete_my_account) sit in the first band of the ⋯ menu, unmarked, while only the console's injected 删除 reaches the red band behind the separator.
What changes
- The destructive test on the header path recognises the spec's
danger(and keepsdestructivefor host chrome andsys_delete), so an app's danger actions land in the last band, behind the separator, painted red — the band that already exists. Prefer one normalisation where the header actions are memoized over a second spelling test, so the inline button, the overflow row and the disabled-reason branch agree. - Nothing else changes:
order,maxVisible,record_morerouting, the console band andcomponent: 'action:menu'pinning stay as they are.
First measurement (before any PR)
On the local stack, a record whose object declares a variant: 'danger' action folded into ⋯ (a fixture, or sys_user's delete_my_account on your own record): confirm where the row sits today, then after the fix, screenshot each. A premise_still_valid: false report (the row is already red and last) closes this card not_planned with the screenshot.
Acceptance
- Test in
page-header-actions.test.tsx(or the overflow-band test of objectui#12529): adangeraction is last, behind a separator, with the destructive row class; asecondaryaction stays in band 1; the console's 删除 unchanged. - Changeset: patch on
@object-ui/components.
Activity
objectstack-fleet commentedon Oct 18, 2026
Claim: PM loop round 8 (
priority:p2)Session:
session_01XVp19E1MYUTFM3dnkwz8gBAccount:
baozhoutao(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-12961-header-danger-bandWorktree:
objectui-issue-12961Domain:
domain:uiSeat:
domain:ui#2File surface:
packages/components/src/renderers/layout/containers.tsx(the header-actions path:isDestructiveActionat about :2039 and its readers, or one normalisation where the header actions are memoized) and tests beside it (page-header-actions/ the objectui#12529 overflow-band test). Plus one.changeset/12961-*.md(@object-ui/components: patch). ⛔order,maxVisible,record_morerouting, the console band andaction:menupinning unchanged; ⛔ no spec key. Stop on breach; explain in the report.Container & model:
S,mode:subagent,model: opus(default tier;dispatch-gates --tier --repo objectstack-ai/objectuireads no path-derived mandate)Clause-②: no
Responsibility:
@object-ui/components' header path testsvariant === 'destructive', the Button primitive's word, where the spec'sActionDefinition.variantsaysdanger| the platform already mapsdanger→destructiveforDeclaredActionsBar's own buttons (DeclaredActionsBar.tsx:409), not on the header path | every record page whose object declares avariant: 'danger'action folded into ⋯ (cloud'ssys_environmentarchive / delete,sys_user'sdelete_my_account).Thread-read: none
Serial constraints cleared: no open objectui PR touches
packages/components/src/renderers/layout/; the in-flight claims (objectui#12954, #12957, #12960, #12894, #12725) name no file underpackages/components/.Claimed at: 2026-10-18T12:24Z
objectstack-fleet commentedon Oct 18, 2026
os-dev-report
{
"issue": 12961,
"status": "done",
"branch": "claude/issue-12961-header-danger-band",
"pr": null,
"session": "session_01XVp19E1MYUTFM3dnkwz8gB — the PM's session (mode:subagent)",
"premise_still_valid": false,
"summary": "Premise is dead: objectui#12764 (PR #12798, commit 69a408f, merged 2026-10-16T21:28Z) already made page:header read the spec's variant 'danger' as destructive, after the card's source read at a8560b2. On origin/main 7f9cc37 the ONE predicate in PageHeaderActions is
isDestructiveAction = (a) => a?.variant === 'danger' || a?.name === 'sys_delete', and it feeds the primary-candidate filter, the three overflow bands, the separator index, the inline button variant and the overflow row's destructive class / disabled-reason branch. A2: there is one copy only; the predicate lives inside the exported PageHeaderActions (objectui#12906) and PageHeaderRenderer renders that component, so the print chrome row and page:header share it. No code written, no PR opened; the empty branch was pushed as the claim marker and can be deleted. Recommend closing #12961 as not_planned, citing #12798 and the measurement below.","tests": "A1, measured on origin/main 7f9cc37 before any edit, with a scratch rendered PageHeaderRenderer test (deleted afterwards, never committed) under os-verify-lock (VERDICT command-exit 1 only because my scratch expectation guessed the order inside the band wrong). Actions: Alpha, Archive Environment (variant danger), Secondary One (variant secondary), Delete My Account (variant danger, locations record_more), maxVisible 1, host Edit/Share/sys_delete as RecordDetailView injects them. Printed reading:
MEASURE row=[\"Alpha\",\"Edit\"] overflow=[{\"label\":\"Secondary One\",\"red\":false},{\"label\":\"Share\",\"red\":false},\"---\",{\"label\":\"Archive Environment\",\"red\":true},{\"label\":\"Delete\",\"red\":true},{\"label\":\"Delete My Account\",\"red\":true}]. So the danger rows (record_header folded and record_more alike) are already last, behind the separator and red, and secondary stays in band 1. That matches the role file's stop condition. In the same run, the existing pin packages/components/src/tests/page-header-danger-12764.test.tsx passed 4/4. It already covers pin (a) folded danger is last and red behind the separator, (c) the CONTROL test with sys_delete unchanged, and (d) inline danger draws bg-destructive and is never the filled primary. Pin (b) is covered by that file's CONTROL band order. Ancestry:git merge-base --is-ancestor 69a408f646 a8560b228527exit 1 (the fix is not in the card's read), and the control leg--is-ancestor a8560b228527 HEADexit 0. Same check for 69a408f against HEAD: exit 0. The repo is not shallow. No ablation, gates or changeset: there is no diff.","mcp_calls": "0",
"api_writes": "1 — POST repos//issues/12961/comments (this os-dev-report). git push of the empty branch is not a REST write.",
"open_questions": [],
"out_of_scope_findings": [],
"gates": "none run: no files changed",
"files_changed": [],
"deviations": "Did not open a PR or write a changeset, by the premise stop rule. The card asked to keep
destructivecounting for host chrome, but on main the predicate deliberately does not readvariant: 'destructive'(#12798 retired it as a non-spec value). The host Delete still counts by the name sys_delete, so the console delete row is unchanged (the measured last red row). This is recorded here and not changed."}
objectstack-fleet commentedon Oct 18, 2026
Closed
not_planned· premise disproved ·domain:uiseat 2 ·session_01XVp19E1MYUTFM3dnkwz8gB· 2026-10-18T12:28ZThe card's own rule applies: "a
premise_still_valid: falsereport closes this cardnot_planned". The dev's measurement is in os-dev-report6249907502; this seat re-verified it onmain.69a408f646on 2026-10-16. Onmain(7f9cc37),packages/components/src/renderers/layout/containers.tsx:2039–:2040 holds one predicate,isDestructiveAction = (a) => a?.variant === 'danger' || a?.name === 'sys_delete', inside the exportedPageHeaderActions. Every reader uses it: the primary pick, the three overflow bands, the separator, the inline button face, and the overflow row's destructive class and disabled-reason branch.page:headerand the print chrome row share it.69a408f646is not an ancestor of the card's read shaa8560b2(git merge-base --is-ancestorexits 1), so the card read a tree from before the fix.PageHeaderRendererwith a foldedvariant: 'danger'action, arecord_moredanger action, asecondaryaction and the host Edit / Share /sys_deletedrew the overflow asSecondary One,Share, separator,Archive Environment(red),Delete(red),Delete My Account(red). So danger rows are last, behind the separator and red, and the secondary action stays in band 1. The existingpage-header-danger-12764.test.tsx(4 / 4 pass) already pins the card's Acceptance (a)–(d).variant: 'destructive', because PR fix(components): page:header reads the spec's variant 'danger' as destructive (objectui#12764) #12798 retired it as a non-spec value. The host Delete still counts through its namesys_delete, so the console's delete row is unchanged.No PR, no code, no changeset. The empty claim branch
claude/issue-12961-header-danger-bandholds no commits. For objectstack#24277 ruling E1: the menu already honoursdangeron any console build whose objectui is at or past69a408f646. Reopen if a console build at that point still draws a danger row in band 1.Generated by Claude Code