You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Sweep card (maintainer-directed, 2026-08-07). One claim, one PR, five Fixes, per-item checklist review — the #6243 / PR #6288 pattern, adopted here to amortize the claim/CI/acceptance overhead that domain:spec currently pays five times over.
Admission criterion (all five members hit the same one)
A surface declared in packages/spec with zero runtime consumers repo-wide, whose disposition under ADR-0049 enforce-or-remove is remove.
Same route for every member: spec-property-retirement playbook — retiredKey() tombstone, ADR-0087 conversion/registry entry, regenerated baselines, docs regen, changeset. That shared machinery is exactly what makes one PR cheaper than five.
Members — verify, then retire
Per-item gate, mandatory before touching any member: re-run the zero-consumer measurement on origin/main (quoted exact names, declaration-site grep per the scan discipline — a bare-name grep hits surviving relatives). If a live consumer turns up, drop that member from the sweep and report it — a wrong retirement is far more expensive than a smaller sweep.
⚠️Measurement first: this card's own hold reason was that response/capability vocabularies may be a host-implementer reference surface. Confirm plugin-hono-server neither implements nor reports them; if it does, this member leaves the sweep and becomes an adopt/conformance card.
no execution-side consumer in this repo (spec + generated docs only)
⚠️SYNC_ARCHITECTURE.md currently recommends it as the destination after the L1 retirement — that prose must be corrected in the same PR, or the retirement contradicts the doc.
Scope discipline
Zero changes beyond the five members. The PR self-certifies: git diff --stat file count must correspond line-for-line to the checklist. This is the line between a sweep and a rider.
Out-of-scope findings are filed separately, as always — PD chore: version packages #10 is not suspended by bundling.
Each retirement keeps its own changeset entry (one file per change; they never conflict).
Reviewed by the checklist, not by diff order: one line per member — landing site | what was declared | what the tombstone now says — plus the consumer-measurement evidence for each. A member without its measurement quoted is not reviewable.
Filed unassigned. Member issues keep pm:queue but are excluded from batch selection while this card is open (double-dispatch guard); all five close together when the PR merges.
Sweep card (maintainer-directed, 2026-08-07). One claim, one PR, five
Fixes, per-item checklist review — the #6243 / PR #6288 pattern, adopted here to amortize the claim/CI/acceptance overhead thatdomain:speccurrently pays five times over.Admission criterion (all five members hit the same one)
Same route for every member:
spec-property-retirementplaybook —retiredKey()tombstone, ADR-0087 conversion/registry entry, regenerated baselines, docs regen, changeset. That shared machinery is exactly what makes one PR cheaper than five.Members — verify, then retire
Per-item gate, mandatory before touching any member: re-run the zero-consumer measurement on
origin/main(quoted exact names, declaration-site grep per the scan discipline — a bare-name grep hits surviving relatives). If a live consumer turns up, drop that member from the sweep and report it — a wrong retirement is far more expensive than a smaller sweep.ServerEvent*/ServerCapabilities/ServerStatusinsystem/http-server.zod.tsplugin-hono-serverneither implements nor reports them; if it does, this member leaves the sweep and becomes an adopt/conformance card.ActionDescriptor.supportsPause/isAsyncresult.suspendlistViews/getView/createView/updateView/deleteView)TestSuiteSchema/TestScenarioSchema/TestStepSchema/TestActionSchema/TestAssertionSchema(qa/testing.zod.ts)ETLPipelineSYNC_ARCHITECTURE.mdcurrently recommends it as the destination after the L1 retirement — that prose must be corrected in the same PR, or the retirement contradicts the doc.Scope discipline
git diff --statfile count must correspond line-for-line to the checklist. This is the line between a sweep and a rider.Landing constraints
packages/specgenerated baselines ⇒ themerge=os-regenfour-step序 applies on every sync:git merge origin/main→git checkout origin/main -- <generated>→ commit the merge first, then regenerate wholesale → assert sibling entries survive. Never a text merge, nevergen:schemawhile in MERGE state (os-regen 驱动指示的gen:schema在 merge 未 commit 时运行,会把 authorable-surface 锚点倒退回旧 merge-base —— 生成器写入、门全绿、静默撤销 main 的锚点推进 #5370).pnpm --filter @objectstack/spec check:authorable-surfacegreen —baseRevis allowed to lag; ⛔ never hand-edit the anchor.Review shape
Reviewed by the checklist, not by diff order: one line per member — landing site | what was declared | what the tombstone now says — plus the consumer-measurement evidence for each. A member without its measurement quoted is not reviewable.
Filed unassigned. Member issues keep
pm:queuebut are excluded from batch selection while this card is open (double-dispatch guard); all five close together when the PR merges.