spec(changes): delete the committed spec-changes.json and its merge=os-regen route; both projections are gitignored and generated at publish only (#22449 B′, card ③)

[objectstack-fleet]

Fixes #22485
Clause-②: no

The committed packages/spec/spec-changes.json and its merge=os-regen route are deleted, and both ADR-0087 D4 projections are gitignored inside @objectstack/spec. This finishes ruling 6078203801 on #22449 (B′): "The committed copies and their two merge=os-regen routes are deleted; check:spec-changes and check:upgrade-guide stop comparing a committed copy." The checks stopped comparing in #22533 (#22482). The guide's copy and route went in #22556 (#22483). This PR removes what was left: the spec-changes.json copy, its route, and every reader of it.

What ships does not change. The publish lane still writes both projections into the package right before the tarball is packed, files[] still lists them, and --verify still refuses a tarball that is missing one. Proof is in the dry run below.

What changed

• The copy is gone. packages/spec/spec-changes.json is deleted from git. .gitignore now lists it and packages/spec/protocol-upgrade-guide.md, which is spec(changes): generate the per-major spec-changes section and the protocol upgrade guide at publish; the pull request still generates both in memory and renders the diff (#22449 B′, condition 1) #22482's note.
• The route is gone. .gitattributes no longer routes either projection. scripts/regen-artifacts.mjs drops the row and records gen:spec-changes in NOT_DRIVER_MANAGED as untracked output. The merge-driver self-test holds that entry against git. check:merge-driver now reads 16 routed paths and 5 untracked dispositions. The comment counts in scripts/git-merge-regen.mjs now read 16 paths and 16 rows, 14 of them defaulted.
• The generator writes no committed copy. build-spec-changes.ts with no flag writes packages/spec/spec-changes.json, which is now untracked and gitignored. It is the same file release-spec-changes.sh --prepare / --generate write for files[] to ship, and the one gen:spec-changes writes locally. --check and --out are unchanged. In lib/projection-cli.ts, committedPath becomes defaultPath. build() is byte-for-byte untouched, so the generated bytes are unchanged.
  • Why not make a bare run a usage error, as docs(spec): the protocol upgrade guide gets a public address, and docs/protocol-upgrade-guide.md stays as a committed pointer stub so the published pointers keep resolving (#22449 B′, condition 2) #22483 did for the guide? Three callers run the no-flag mode and read its output. release-spec-changes.sh calls it at --generate, at the first-ever --prepare, and at --prepare --previous-package. Cloud's changed-spec-surface guard runs pnpm run gen:spec-changes in its pinned framework checkout and reads only that output (cloud#2750, comment 6083243420 on spec(changes): generate the per-major spec-changes section and the protocol upgrade guide at publish; the pull request still generates both in memory and renders the diff (#22449 B′, condition 1) #22482). A usage error there would turn cloud's only required check red on its next pin move, and would also need an edit to the release script, which is outside this card's file surface. The guide's no-flag mode moved to a different target. This one keeps its target, and the target is no longer tracked.
• H2: the ADR-0087 gate's parser-rot witness is generated. See below.
• H3: an unrouted pending marker now prints a remedy that works. See below.
• check-generated.ts's row now names the gitignored copy. os-regen-merge.sh gains a note that its 2026-09-17 readings were taken over a path that has since left the route list. That script reads the routed list from .gitattributes at run time and hard-codes none.
• Rider declared by the seat (pointer 6094908270, carried from the spec: ListView.userActions.editInline defaults to true in v18 and editInline: false is the opt-out; objectui reads an absent key as on (maintainer ruling 2026-10-10, after objectui#5144) #22605 dev): packages/spec/src/migrations/entries/README.md, under "What this does not fix", told authors that both projections "still have to be regenerated and committed when an entry lands". It now says what is true after the deletion. An entry regenerates registry.ts alone, with gen:migration-registry (gate: check:migration-registry). The two projections are generated at publish, and in memory on the pull request. ⛔ Neither is regenerated for an entry. The spec-changes.json and protocol-upgrade-guide.md are unsharded generated artifacts whose merge safety rests on a LOCAL-only git driver — the merge queue rebuilds server-side, where no custom driver runs #8344 merge-queue measurement stays, marked as history kept for its registry.ts conclusion.
• Forced, outside the claim's file list:
  • scripts/check-future-spec-major.mjs deletes its QUOTATION_EXEMPTIONS row for the deleted copy. Measured: HEAD's version of the gate, run on this tree, exits 1 with QUOTATION_EXEMPTIONS entry 5 (packages/spec/spec-changes.json, major 4997) matched NOTHING. After the deletion the gate exits 0 with 6 witnessed ledger entr(y|ies), every witness still matching.
  • scripts/objectui-changeset-digest.mjs: its fw-gate sandbox runs a copy of the real ADR-0087 gate, so it now stages the stand-in witness generator instead of a fixture spec-changes.json. Without that change, its round-trip cases die at the input assertions.

H1: every reader of the committed copy, and what it reads now

Reader	After this PR
scripts/check-adr-0087-registration.mjs (assertInputs, formerly :2184)	Generates the projection at the rev it judges (H2). It no longer reads any copy.
packages/spec/scripts/build-spec-changes.ts no-flag mode	Writes the untracked, gitignored package copy (above).
packages/spec/scripts/check-generated.ts	Runs check:spec-changes in memory. That check never goes stale, so --fix never regenerates. The row text names the gitignored copy.
scripts/regen-artifacts.mjs, .gitattributes, scripts/git-merge-regen.mjs, .githooks/pre-commit → check-regen-pending.mjs	No row and no route. The driver never sees the path, and an old marker naming it gets the UNROUTED refusal (H3).
scripts/pm/os-regen-merge.sh	Reads routes at run time (16). It has no hard-coded list. Only the prose note changed.
scripts/release-spec-changes.sh (--prepare / --generate / --verify / --attach)	Unchanged. It writes into the package and verifies the packed copy (dry run below).
packages/spec/package.json files[]	Unchanged. It ships what the lane wrote.
scripts/check-release-spec-changes.mjs	Unchanged. It reads the packed tarball and a fresh generation, never the tree copy.
scripts/release-verify-npm.mjs (MANIFEST at :2063)	A path inside its stubbed workflow fixture. Unchanged, and its self-test is green.
packages/spec/scripts/render-projection-diff.ts (legacyPath)	Generates both sides with --out. The legacy path is read only inside a throwaway archive of a base that predates --out, never from this checkout. Unchanged.
packages/cli/src/utils/spec-release-changes.ts	Reads the installed package's release section. The committed copy never had one, so it answered null in the workspace before this PR, and an absent file answers null too. Unchanged.
packages/spec/src/**/*-retirement.test.ts (EXCLUDED_PREFIXES)	Walk the disk and still exclude the path, which is still right: a local gen:spec-changes output sits there, gitignored. Unchanged.
scripts/check-future-spec-major.mjs	Scans tracked files. Its row for the copy is removed (forced, above).
scripts/check-regen-pending.mjs self-test fixture	Used the path as "a real REGEN_ARTIFACTS entry". It now derives its row from the table, and uses the path as the retired-route scenario.
scripts/objectui-changeset-digest.mjs self-test sandbox	Staged a fixture copy for the gate. It now stages the stand-in generator (forced, above).
cloud scripts/audit-spec-changes.mjs	Reads gen:spec-changes output in its pinned checkout since cloud#2750. That output path is unchanged here.

H2: the witness is generated at the judged rev

Before this PR, assertInputs read HEAD:packages/spec/spec-changes.json as its independent witness that extractIds() still sees every id the build sees. Since #22533 that copy has been aging. A withdrawn migration id stays in an aged copy and false-reds the gate, and deleting the copy reds it outright.

generatedWitness() now lays out the judged rev's packages/spec/{src,scripts,package.json} with git archive. It borrows the checkout's node_modules and runs build-spec-changes.ts --out under the installed tsx. So the witness describes exactly the rev being judged, never the working tree. A missing runner, a missing generator, a generator that exits non-zero, or one that writes nothing is each a red input, never a skip. On this tree the real run generates 414 ids, all of them seen by the parser. The gate run is about 2 s.

• Pins. Battery W1-W6, with 9 cases, drives the real generatedWitness over a stand-in generator. The stand-in evaluates the fixture registry under tsx, as the real one does.
  • W1: a withdrawn id with a stale committed copy at HEAD is green, and the witness the gate used holds only the kept id.
  • W2: an uncommitted edit stays green, and the same edit committed is parser drift that names the id.
  • W3 to W6: a missing generator, a failing generator, a silent generator, and no tsx are each red.
  • I1 (parser rot) and the I2 entry-point run now go through the real path too. The self-test reports 450 assertions, up from 441.
• Ablation, via scripts/ablation-replace.mjs at 538bec7e1d. The witness read was put back to the committed-copy read.
  • The self-test exits 1 with 14 failures, including W1: a withdrawn migration id must NOT red the gate -- got: ledger parser drift … - withdrawn-one.
  • Restored: blob 7e383d48ea90 == HEAD, and git diff HEAD is empty.

H3: a pending path whose route was retired

The premise holds for the class, but not for this card's own path. The deletion does not reach the stuck marker: git runs no merge driver on a modify/delete, so a branch carrying spec-changes.json writes no marker for it (H4). The stuck marker comes from the guide's route, retired by #22483, and from any marker an earlier merge left.

Measured with the real driver and hooks, in a detached scratch worktree:

• The scratch worktree starts at 874a38d8d9, which is 514bf3c101^1, before docs(spec): the protocol upgrade guide gets a public address, and docs/protocol-upgrade-guide.md stays as a committed pointer stub so the published pointers keep resolving (#22449 B′, condition 2) #22483. A commit there edits docs/protocol-upgrade-guide.md. It then merges this branch (cec190b908).
• The merge auto-completes with exit 0. The marker holds docs/protocol-upgrade-guide.md, and the tree holds the branch's generated guide: the stub was dropped.
• The next commit exits 1:

  ✗ docs/protocol-upgrade-guide.md — UNROUTED: pending, but scripts/regen-artifacts.mjs has no row for it any more, so no gate
      can verify it and nothing clears it by itself. The merge kept THIS side of it whole and dropped
      the side it merged in: compare them (`git diff HEAD^2 -- docs/protocol-upgrade-guide.md`), keep or restore the
      right bytes, then release it:
      node scripts/check-regen-pending.mjs --release docs/protocol-upgrade-guide.md
  

• Following it:
  • git diff HEAD^2 shows +1590 −9.
  • git checkout HEAD^2 -- docs/protocol-upgrade-guide.md restores the 15-line stub.
  • --release exits 0 with "nothing left pending, marker cleared".
  • The commit exits 0, and --pre-push exits 0.
  • The guide is byte-identical to this branch's.
• Control: 18d999031b's check-regen-pending.mjs, run on the same marker, exits 1 with recorded as pending but absent from scripts/regen-artifacts.mjs (cannot verify) and This check clears itself the moment they are current — nothing to reset by hand.. It has no --release, and its run leaves the marker in place.

The path stays BLOCKED, never passed. Passing it would land exactly the drop the driver made, and #22556 recorded the refusal as the one thing that keeps that drop loud. --release PATH refuses a path that still has a row, because its gate discharges it, never a hand. It also refuses a path the marker does not hold, and keeps every other line and the deferral record.

• Pins (fixtureSelfTest, in the script's own self-test harness):
  • The scenario's premise is asserted: the path has no row.
  • The refusal is exit 1 and no longer claims "nothing to reset by hand".
  • The printed command is parsed and run as printed, with no fallback spelling. It clears the marker, and the next push is accepted.
  • --release refuses a routed path and an absent path, and leaves a routed path's debt and its stale refusal intact.
  • The fixture's pending row is now derived from REGEN_ARTIFACTS, so retiring a row cannot break the fixture again.
• Ablation, via scripts/ablation-replace.mjs at 538bec7e1d. The printed --release line was dropped.
  • The self-test exits 1: "printing a release command", "the push is refused with the same remedy", "THE PRINTED REMEDY WORKS" and "the next push is accepted" go red, both directly and in the ambient-git-env inner run.
  • Restored: blob ca127a866a97 == HEAD, and git diff HEAD is empty.

H4: a branch that still carries a regenerated copy

What the author does:

git rm packages/spec/spec-changes.json && git commit --no-edit

The conflict is modify/delete, so git runs no merge driver and leaves no marker. The deletion is the resolution, and nothing regenerates the copy in git any more. Measured on a scratch commit at 18d999031b that edited the copy, merging cec190b908:

• the merge exits 1 with CONFLICT (modify/delete): packages/spec/spec-changes.json deleted in cec190b908… and modified in HEAD, and no marker is written;
• the command above exits 0, and the resulting tree equals this branch.

PR #22421 (#11509, seat 2) is such a branch today. The same line is in .gitattributes and in NOT_DRIVER_MANAGED's reason, which is where a merging author meets the route list.

Publish lane dry run (no release act)

At 928bee8573, with RELEASE_VERSION=17.7.0:

• --generate exits 0 and writes both files into packages/spec/. git status --short stays empty: --ignored shows both as !!.
• --verify exits 0 with ✓ registry projections verified against a fresh generation: spec-changes.json (2 per-major record(s), 121 converted / 414 migrated) and protocol-upgrade-guide.md (1617979 bytes) match the artifact. and ✓ no per-release section is owed: this lane (--generate) publishes none..
• The packed objectstack-spec-17.7.0.tgz lists package/spec-changes.json and package/protocol-upgrade-guide.md. Its spec-changes.json is byte-identical to the generated one (cmp).
• Control: with the generated spec-changes.json removed, --verify exits 1 with the artifact about to publish ships no spec-changes.json. A consumer's node_modules cannot be handed a tarball without it.

Changeset: none, skip-changeset

No published byte changes:

• build() is untouched, so the lane generates the same bytes;
• packages/spec/scripts/** is not in files[];
• the deleted copy never shipped as-is, because the lane overwrites it before packing (spec(changes): generate the per-major spec-changes section and the upgrade guide at publish; the pull request generates both in memory and renders the diff (#22449 B′, condition 1) #22533).

So no package warrants a bump, and this PR takes skip-changeset. The dispatch asked for H4's line to go in a changeset. With no changeset, that line is in H4 above and in the two in-repo places named there.

Governed lines that name the copies or their routes (not edited here; the skills lane carries them)

• docs/adr/0087-metadata-protocol-upgrade-contract.md:263–:266: "the deletion of the committed copies (the per-major projection in packages/spec/spec-changes.json, the guide body at docs/protocol-upgrade-guide.md) and of their two merge=os-regen routes in .gitattributes — spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485, which lands last." It is written as not yet landed.
• docs/adr/0087-metadata-protocol-upgrade-contract.md:268–:270: "Until spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485 lands the committed copies stay in the tree, and the generators' no-flag mode (gen:spec-changes, gen:upgrade-guide) still writes them, but check:spec-changes and check:upgrade-guide no longer compare them". The guide half was already stale since docs(spec): the protocol upgrade guide's public address is one docs-site page per protocol major; docs/protocol-upgrade-guide.md becomes a pointer stub (#22449 B′, condition 2) #22556.
• docs/adr/0087-metadata-protocol-upgrade-contract.md:283–:284: "The 2026-08-06 note above ("deliberately unsharded") is about that committed copy, which spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485 removes."
• docs/adr/0087-metadata-protocol-upgrade-contract.md:207–:208: "spec-changes.json is deliberately unsharded — it is keyed by version, which has never been a conflict surface." This is about the committed copy, which no longer exists.
• docs/adr/0087-metadata-protocol-upgrade-contract.md:438–:440: "Those copies stay in the tree until spec(changes): delete the committed spec-changes per-major projection and the upgrade guide copy, with their two merge=os-regen routes, once generation at publish has landed (#22449 B′) #22485 deletes them, and check:spec-changes / check:upgrade-guide no longer compare them."
• docs/adr/0087-metadata-protocol-upgrade-contract.md:219–:220: the ruling's own words, quoted. They are listed only because they name the routes, and they are not to be edited.
• AGENTS.md:717: "| An ADR-0087 conversion / migration registry | check:spec-changes, check:upgrade-guide | gen:spec-changes, gen:upgrade-guide |". Neither check can go stale any more, and both generators write only untracked files. The row that holds for an entry is check:migration-registry / gen:migration-registry.
• AGENTS.md:706: "packages/spec has eight checked-in generated artifacts, each with its own CI gate." The two projections are no longer checked in, so the count needs a re-measure.
• .claude/skills/spec-property-retirement/SKILL.md:260: "然后 gen:spec-changes、gen:upgrade-guide、gen:api-surface、gen:docs。"
• .claude/skills/spec-property-retirement/SKILL.md:336–:337: "所以纯散文改动也要 gen:spec-changes + gen:upgrade-guide,和任何别的改动一样。"
• docs/adr/0120-unique-scope-vocabulary-and-null-safe-tenant-uniqueness.md:459–:460: "regenerate with the conversion entry (gen:spec-changes, gen:upgrade-guide)." This is a record of what that PR did, likely left as history.

Measured and not stale: .claude/skills/pm-dispatch/SKILL.md:296, the hot-file row for os-regen-merge.sh, which names no path, and references/landing-operations.md:38, which reads .gitattributes live.

Verification

Gate union read at cb1ac74e76, the final head. It is the 87 commands the dispatch named, plus what node scripts/pm/dispatch-gates.mjs --commands derives on this diff (105 today), 110 commands in all. Each ran with its exit code captured before any pipe, and all 110 exited 0. --ran reports Run reconciliation — 105 derived, 105 run, 0 NOT-MEASURED, 0 UNRUN. That includes:

• check:merge-driver: 16 routed paths, and 5 untracked disposition(s) still hold — git tracks none of those paths;
• check:changeset-gate-self-tests and the ADR-0087 gate itself (--base origin/main: green, with a generated witness);
• check:generated: ✓ All 15 generated artifacts are up to date, with check:spec-changes spec-changes.json (gitignored; written into the package at publish);
• check:objectui-changeset, check:pm-dispatch-gates, check:type-check-debt and check:nul-bytes.

Under scripts/pm/os-verify-lock.sh (VERDICT command-exit 0, at 928bee8573; the README rider after it touched no test input):

• @objectstack/spec: build exit 0; typecheck exit 0, and tsc -p tsconfig.scripts.json --listFiles lists all four touched script and test files;
• vitest --project local: 642 files, 19146 passed, 1 todo;
• --project repo: 54 files, 915 passed;
• release-verify-npm.mjs --self-test: 115 cases across 13 batteries;
• check-release-spec-changes.mjs --self-test: 31 batteries.

Acceptance notes (not filed)

• Size. About 8,000 changed lines, 7,272 of them the deleted generated JSON. That is over the 3,000-line threshold, so this lands the Tier H way: an authorized approval or a human merge.
• Landing rule A relaxation. The maintainer's temporary relaxation for D3-entry PRs (6091886885, corrected by 6091996837) ends when this lands.
• .github/workflows/lint.yml:4728 lists spec-changes.json among the artifacts merge=os-regen protects. Comment only. carrier: none.
• .github/workflows/pr-automation.yml:959–:961: "the two scripts import only node builtins and repo-local modules, so a failed install cannot change their verdict". The import half is still true. The ADR-0087 gate now needs the installed tsx for its witness, so a failed install also reds that step, in the enforcing direction. Comment only. carrier: none.
• scripts/check-published-files.mjs:291: the spec-changes.json registration reason could say "generated at publish (it is not in the tree)", as the guide's does. carrier: none.
• docs/spec-generated-artifact-sharding.md:42 describes the committed file's merge shape as current. Design history. carrier: none.
• Cost. check-adr-0087-registration --self-test costs about 7 s more on this shared box (20.7 s → 27.5 s): 8 real generator spawns.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

⚠️ 2 changed file(s) yielded no anchor (packages/spec/spec-changes.json, packages/spec/src/migrations/entries/README.md), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files. Nothing else in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 1 changed package(s)).

What this run could not see

• 2 changed file(s) yielded no anchor (packages/spec/spec-changes.json, packages/spec/src/migrations/entries/README.md) — pages documenting those are invisible to this run
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 139 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 1b99388505d33c940757e85685ca925d6cd57f04 → packageMentionDocs.

[objectstack-fleet]

维护者速读

domain:spec seat 1 (#6017) · os-tesla · session session_01VZqqwTj2wsihZEbfT6yyYN · 2026-10-10T08:18Z。待批的是 head cb1ac74e76。席位复核 ACCEPT 记录是 #22485 的 6095601210;CI 39 项:32 通过,7 项按预期跳过(已核对跳过名单),0 失败。本 PR 不碰受管面,走人工合并只是因为体量:+591 / −7447,其中 7272 行是被删除的生成文件 spec-changes.json,超过 3000 行门槛。

• 改了什么: 落实决策卡 [Decision] generated protocol-change records (spec-changes.json per-major section, docs/protocol-upgrade-guide.md): keep committing them, or generate them at publish like the release section #22449 的 B′ 第三步,也是最后一步。
  • 把仓库里提交着的生成文件 packages/spec/spec-changes.json 删掉,并加入 .gitignore;
  • 去掉它在 .gitattributes 里的 merge=os-regen 合并路由,以及重新生成表里的那一行。
  • 发版时仍由发版流水线生成这份文件、放进 npm 包,包里内容不变。PR 阶段在内存里生成并检查。
  • 顺手修了三处门禁和脚本:
    • ADR-0087 登记检查原来读仓里那份文件做"解析器自检",现在改为在被检版本上现场生成;
    • 合并后卡住的"待重新生成"标记,现在会打印一条能用的解决命令(--release);
    • 迁移条目 README 里"每加一条都要重新生成这两个文件"的过时说明已改正。
• 为什么改: 这份文件每加一条迁移条目都要重新生成并提交,是并行 PR 互相冲突、反复合并的主要来源;今天就有三个 dev 白白生成又撤回。你在 [Decision] generated protocol-change records (spec-changes.json per-major section, docs/protocol-upgrade-guide.md): keep committing them, or generate them at publish like the release section #22449 裁定 B′:只在发布时生成。本 PR 删掉最后一份提交着的副本。
• 风险与代价(含回滚):
  • 发布出去的包内容零变化。dev 用 17.7.0 演练了发版流水线:生成、校验都通过,打出的包里两份文件与现场生成逐字节一致。
  • 还带着旧副本的分支,合并 main 时会遇到"修改/删除"冲突,解决办法是 git rm 这个文件(已写进 .gitattributes 注释)。PR feat(spec)!: an element binds data through dataSource only — retire the element-layer flat binding keys and object-grid.defaultFilters (#11509) #22421 已按此预先安排。
  • 回滚就是还原这个 PR。
  • 你之前给的放宽规则(带迁移条目的 PR 显示可合并就直接入队)在本 PR 落地后结束。
• 席位意见: 建议批准。
  • 我逐段读过 diff:删除干净,发版路径保留,门禁只加强、没有放松。
  • 新增的 --release 只能释放"已没有生成规则"的路径,对仍有规则的路径和不在待办里的路径都会拒绝,并有自测和消融验证。
  • 不写 changeset 是对的:没有任何发布字节变化。
  • 还有几处受管面文字需要同步(ADR-0087 进度说明、AGENTS.md 的重新生成表和"八个生成物"计数、spec-property-retirement SKILL.md),落地后我立一张 skills 车道卡,不进本 PR。
• 你要做的: 在本 PR 点 Approve。批准后由本席走合并队列落地,你不用手动合并。