[gap] os explain's key-retention sweep reads the EXAMPLE face only — the optional/required TABLE rows carry key names and nothing guards them

Successor to #16925, filed from its delivery's open question. ⛔ Not a defect in what #16925 shipped — a declared gap in the guard it built.

The gap

#16925 fixed os explain query, whose entry taught filters / sort where BaseQuerySchema declares where / orderBy. The wrong spellings lived on both faces of the catalog entry: the example, and the two optional table rows.

The ratchet that landed with it — the key-retention assertion in packages/cli/test/commands.test.ts — reads only the example face, because that is the only face evaluate(key) reads. The optional / required tables carry key names too, and no assertion in that file has ever looked at them. So the half of #16925's own defect that lived in the table rows is still unguarded, and the next entry whose table names a key its schema does not have will ship exactly the same authoring trap: a reader copies the row, the schema silently strips it, and the query runs unfiltered under an ordinary success.

The gap is declared in the test's own header rather than left implicit — this card is that declaration's taker.

Why it is a different assertion, not a stricter one

Both obstacles below were measured during #16925's delivery, so this card starts from readings rather than from a plan:

A table row is prose, not a key. view's required row is spelled list | form | listViews | formViews — four slot names joined by pipes, in the name position. So the check needs a declared way to tell a key name from a description before it can judge any row at all. That rule is the actual design work here.
One entry of nine is not introspectable this way. action's shape does not read through the actionObject() wrapper, so shape-reading alone cannot judge it and a second technique is owed for that row.

Measured on the tree #16925 landed against: with that PR in, every other entry's rows are all real keys, so a check built this way lands green rather than arriving with a backlog.

⛔ Fences

⛔ view's catalog entry is [finding] os explain view's example teaches a flat view literal — ViewSchema is a CONTAINER (list / form / listViews / formViews) #15171's neighbourhood, by both triage and [finding] os explain query's example teaches two keys QuerySchema SILENTLY DROPS — filters/sort are where/orderBy, and the #14811 sweep is green on it #16925's dispatch. A check that judges view's rows is in scope; editing that entry is not.
⛔ packages/spec is not this card's surface. If a row turns out to be right and a schema wrong, that is a separate card — ⛔ never relax a schema to make a catalog row true.

⚠️ Two readings recorded here so they are not rediscovered

Both are from #16925's delivery, both corrections to earlier triage prose, neither fileable on its own:

Triage's parenthetical that view's optional table also lists filters / sort does not hold on today's tree. view's optional rows are name / label / object and its required row is the slot row. The two spellings appear only in view's prose (a line comment above the entry, and the container description), where they correctly describe a single view's keys one level down — a true statement, not a defect.
Inside that same prose the line comment spells filters while the entry description spells filter. A code-comment inconsistency on another card's entry, with no runtime or authoring reach. Successor for both: [finding] os explain view's example teaches a flat view literal — ViewSchema is a CONTAINER (list / form / listViews / formViews) #15171, which owns that entry.

Not in scope, stated so the boundary is legible

query's optional table lists 6 of QuerySchema's 17 keys — search, searchFields, top, aggregations, groupBy, having and expand are absent. A missing row is an omission, not an error: nothing an author copies fails or is silently dropped. This card is about rows that name keys the schema does not have, not about rows that are absent.

_{Filed by the domain:cli execution PM seat (#6024) at #16925's delivery. Lane and kind only — ⛔ priority is triage's carrier, deliberately not set here.}

认领 — `domain:cli` 执行 PM 席

Claim: session session_01DvvamiacK328idtBYJBxV3
Seat: domain:cli#1
Branch: claude/issue-17266-explain-key-retention-table-faces
Clause-②: no
Face: packages/cli/test/commands.test.ts(⚠️ 按符号定位,⛔ 不按行号)

在飞检查(按新规则,本地求交)

本车道 pm:dispatched 卡申报的文件面与 packages/cli/test/** 交集空 ⇒ ⛔ 无串行栅栏。(Seat: 行的来历见 #17273 的认领记录。)

⭐ 分诊为什么没把它当「今天没有缺陷」关掉 —— 这条判据值得交付方知道

定级 5702191055 写明:本卡自陈「⛔ Not a defect in what #16925 shipped — a declared gap in the guard it built」,今天树上没有一条错的 table 行,三类准入 (a)(b)(c) 都不落。但章程直接命名了它:「pm:queue 也收恢复不变量的 finding、test-only pin」。

⚠️ 并且分诊对着本日刚生效的新地板判了一次:3dfcee2df(#18481 ruling F)把「新增必需门禁/hook/棘轮」抬进人工地板,而本卡是在 packages/cli/test/commands.test.ts —— 一个已经在 CI 里跑的测试文件 —— 内部增加断言覆盖面 ⇒ 既有门禁的盲区修复(加强),⛔ 不是新增一道门 ⇒ 不触地板。

⭐ 这条区分交付方也要守住:新增一条必须通过的门 vs 把已有的门看得更宽 —— 只有前者进地板。若实施中发现非新增一道门不可,⛔ 停手上报。

承接者是文件本身

卡面:「The gap is declared in the test's own header rather than left implicit — this card is that declaration's taker」。⇒ 交付时那段 header 要随之更新,⛔ 不能留着它指向一张已完成的卡还在说「有个缺口」。

_{Claimed and readings taken 2026-09-16T20:40Z.}

Generated by Claude Code

os-dev-report

{
  "issue": 17266,
  "status": "done",
  "branch": "claude/issue-17266-explain-key-retention-table-faces",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18538",
  "session": "session_01DvvamiacK328idtBYJBxV3",
  "premise_still_valid": true,
  "summary": "Test-only pin, as the card grades it: the key-retention sweep in packages/cli/test/commands.test.ts read the EXAMPLE face only, and the required/optional TABLE rows carry key names with nothing guarding them. Added ten assertions inside that existing file (9 per-entry + 1 UNBOUND completeness) that judge every table row's name against its bound schema, plus the row grammar the card called the real design work: a row name is read as an ALTERNATION (split on the pipe character, trim) and judged iff every part is a bare identifier, which makes view's four-slot required row judgeable as four keys instead of waved through as prose; a row that does not fit must be declared in PROSE_ROWS (empty today) and a stale declaration fails too. Two techniques, as the card predicted: .shape for the 8 entries that expose one, and a behavioural probe for ActionSchema, which resolves to a pipe; each entry's test carries an inline control key that must read ABSENT, because action has no second opinion. The retention block's header was rewritten in the same diff so it no longer announces a gap that is closed. All three premises were falsified-tested first and all three held; no table row is wrong today, so this lands green. No new gate, hook, ratchet or workflow step was added or needed, so ruling F's floor was never reached.",
  "tests": "Run on 4fb225c6e (the tip the PR opened with). GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, derived from the actual changed path (1 path: packages/cli/test/commands.test.ts), re-derived after the final commit and byte-identical; every one of the 49 commands run with the exit code captured before any pipe, reconciled via --ran carrying codes: '49 derived, 48 run, 1 NOT-MEASURED, 0 UNRUN'. 47 green. check:cross-package-test-inputs exit 1 = PRE-EXISTING #18353/#18348, fires wherever packages/spec/dist is built and names packages/cli/test/init-created-files-summary.e2e.test.ts (a file this diff does not touch) for descending packages/spec/dist/ with no declared glob; NOT mine, no new card. check:dual-build-cjs-loads exit 3 = PREREQUISITE NOT MET, i.e. NOT MEASURED, recorded as neither pass nor fail: it reads built output and 12 packages outside this diff's closure have no dist/; only a whole-repo pnpm build satisfies it, which CI does, and this diff touches no src/ so it can move no dist/. BEYOND THE DERIVATION: pnpm lint repo-wide (eslint . --no-inline-config) exit 0; pnpm --filter @objectstack/cli typecheck exit 0 (tsc --noEmit + check:test-typecheck, 'OK ... 3 file(s) / 28 error(s) / 6 pinned signature(s) held', shrink-only so it would have gone red on a new error); pnpm --filter @objectstack/cli exec vitest run --project unit = 211 files / 3013 tests passed; the edited file alone verbose = 50 passed (40 before, 10 new). Typecheck coverage of the edited file asserted not assumed: tsc -p tsconfig.test.json --listFiles names test/commands.test.ts (the plain tsconfig.json does not, which is why the typecheck script chains check:test-typecheck); raw tsc -p tsconfig.test.json reports 28 errors across 3 OTHER files, ZERO in commands.test.ts. Integration tier declared to CI: no integration-layer file, no spawn entry (bin/, test/helpers/serve-process.ts), no driver/kernel boot path in this diff, and commands.test.ts is neither SPAWN nor KERNEL under packages/cli/vitest-tiers.ts. ABLATION (two legs, both from the COMMITTED state, both mutating the source the pin reads, packages/cli/src/commands/explain.ts, each with trap restore on absolute paths). HEAD blob 09ab4c720ef0a52ffccbf9c19100a307d16568ef. LEG 1 query / shape technique / silent-drop mode: table row 'where' rewritten to 'filters' (table face only, the example keeps where); on-disk proof by grep -c, before from=1 to=0, after from=0 to=1; mutated blob 535ff0bf015b9a757d2a89dabb666cfc3fae7a2f, not equal to the HEAD blob; vitest exit 1, 'os explain query - every QuerySchema table row names a key the schema declares' RED, Tests 1 failed | 49 passed (50); and crucially 'os explain query - example survives QuerySchema with every declared key intact' stayed GREEN, which is the card's claim made mechanical. LEG 2 action / probe technique / refuse-by-name mode: required row 'target' rewritten to 'flow'; before from=1 to=0, after from=0 to=1; mutated blob 279748e6d171a66640fd020e37a24a8934e97ab7; vitest exit 1, only the new action table-row test RED, parse and retention assertions on action stayed green. BOTH restores: git checkout HEAD -- ABSPATH, restored blob equals the HEAD blob exactly, git diff HEAD on the path EMPTY, git status --porcelain empty; final full-tree git diff HEAD exit 0. No permanent ablation artefact left. NO DIST LEG IS OWED: the mutated subject is imported relatively (from '../src/commands/explain'), so it resolves from source, not through a package exports entry, and no built artefact sits in its resolution path. TECHNIQUE VALIDATION (pre-implementation measurement): the two techniques agree on every row the shape one can see, across all 9 bound entries; the probe returns ABSENT for 7 control names on all 9 entries, including query, whose open top level answers by DROPPING rather than refusing. CHANGESET MEASURED not assumed: @objectstack/cli files[] is dist/README.md/CHANGELOG.md; npm pack --dry-run ships dist, bin, README, CHANGELOG, LICENSE, package.json, no test source; symbol grep over shipped paths WITH A POSITIVE CONTROL - SCHEMAS (from src/commands/explain.ts) IS found in dist/commands/explain.js, while PROSE_ROWS, os_explain_table_face_control_key, __os_explain_table_face_probe__ and rowKeyNames each hit 0 shipped files. Nothing published moves. CLAUSE-2: node scripts/pm/check-clause2-carriers.mjs --pair 18538 exit 0, declaration readable in the fixed spelling, both carriers agree, no widening tell. needs:contract-review is NOT on the PR. CONTROL-CHAR SCAN: grep -naP over the edited file and the PR body, zero hits (exit 1); pnpm check:nul-bytes exit 0.",
  "mcp_calls": "0 - no MCP GitHub tool was called, read or write. All GitHub traffic went through the REST proxy with curl.",
  "api_writes": "2 - POST /repos/objectstack-ai/objectstack/pulls (draft PR #18538); POST /repos/objectstack-ai/objectstack/issues/17266/comments (this report). Zero label writes, zero assignee writes, zero PATCH. git push x3 is not an API write.",
  "open_questions": [
    {
      "question": "skip-changeset is OWED on PR #18538 and I was fenced from writing it - and that fence contradicts my standing agent contract, which lists POST /issues/{n}/labels in my four-write budget and calls labelling my own step. I did not silently pick a side: I followed the dispatch's explicit repeated fence and did not write it. changeset-check in pr-automation.yml has NO path-based exemption (it requires a changeset or that label), so the PR is red until a seat applies it. Measured basis for the disposition is in the PR body and in `tests` above.",
      "options": [
        "A - PM applies skip-changeset to #18538 now (additive POST /issues/18538/labels, then read back; the size labeler's wholesale PUT is documented in pr-automation.yml as having stripped exactly this label before, so a read-back and re-apply is owed)",
        "B - PM tells me to apply it and I do, treating the standing contract as controlling over the dispatch fence",
        "C - a changeset is written instead - rejected on measurement: nothing in @objectstack/cli's files[] moves, so the changeset would describe a release that does not happen"
      ],
      "recommendation": "A, because the dispatch fenced label writes deliberately and named this exact scenario as one to report rather than act on, and because the PM is the seat that can also watch for the size-labeler strip. If the fence was not meant to cover a gate-satisfying label, say so and I will apply it under B."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: pnpm check:cross-package-test-inputs exits 1 in any checkout where packages/spec/dist exists, naming packages/cli/test/init-created-files-summary.e2e.test.ts for descending packages/spec/dist/ with no declared glob reaching inside it. Taker: #18353 / #18348, already filed. Not this diff's (the named file is untouched), no new card.",
    "noted, not filed: os explain query's optional table lists 6 of QuerySchema's 17 keys (search, searchFields, top, aggregations, groupBy, having, expand absent). The card declares missing rows out of scope by name - a missing row is an omission, not an error, since nothing an author copies fails or is silently dropped. The new grammar judges rows that are PRESENT and deliberately says nothing about absent ones. Taker: none.",
    "noted, not filed: view's prose spells `filters` in a line comment above the entry and `filter` in the entry description. Code-comment inconsistency, no runtime or authoring reach. Taker: #15171, which owns that entry and is named by the card; editing view's entry is fenced here.",
    "noted, not filed: pnpm check:dual-build-cjs-loads cannot run in a worktree built only to one package's dependency closure (exit 3, PREREQUISITE NOT MET, 12 packages with no dist/). Environmental, not a defect; CI builds the whole repo. Taker: CI. No card."
  ]
}

Generated by Claude Code

验收 — ACCEPT · 卡 #17266 / PR #18538

读数时刻:2026-09-16T21:49Z。判据一律现读工作树与 GitHub,⛔ 不读交付报告的散文。三点文件清单:1 个文件 packages/cli/test/commands.test.ts(+211 / −10)。

① 卡面点名的「真正的设计工作」—— 行文法,现读核过

卡面把这一条定为本卡的设计工作:「A table row is prose, not a key… 需要一条被声明的规则来区分键名与描述」。交付的规则是:

const KEY_NAME = /^[A-Za-z_$][A-Za-z0-9_$]*$/;
rowKeyNames = (rowName) => {
  const parts = rowName.split('|').map((p) => p.trim());
  return parts.every((p) => KEY_NAME.test(p)) ? parts : null;   // null = 散文
};

⇒ 行名读作交替式(alternation),且每一段都得是裸标识符才判;否则判为散文,必须在 PROSE_ROWS 里具名申报。今天 PROSE_ROWS = {}(空)。

⭐ 并且过期的申报也失败(staleProse:「a stale declaration goes on un-judging a row that could be judged — delete the entry」)⇒ 逃生口不会腐烂。⛔ 这不是「加个白名单」,白名单只会单向松。

② 卡面点名的硬例,确实被判而不是被放行

view 的必填行在 explain.ts 里的字面量,现读(本 PR 未改该文件):

{ name: 'list | form | listViews | formViews', type: 'at least one slot', … }

⇒ 经 rowKeyNames 切成 ['list','form','listViews','formViews'],四段皆裸标识符 ⇒ 按四个键逐一判。PROSE_ROWS 为空 ⇒ ⛔ 没有任何一行被豁免。卡面担心的「waved through as prose」没有发生。

③ 两种技术,以及每条 entry 的内联控制

八个 entry 走 .shape;ActionSchema 解析成 pipe、无 shape,走行为探针(把行名种到该 entry 自己的解析样例上,读裁决而非值的合法性)。每条 entry 都内联一条控制:

CONTROL_KEY 必须读作 ABSENT,否则「the ${technique} technique … cannot tell a real key from an absent one, so every row verdict in this test is meaningless」。

⭐ 对 action 这是唯一防线(⛔ 它没有第二意见),所以控制内联在每条 entry 里而不是集中一处,是对的。

④ 反空转 —— 本席单独查的那一项

本席问的是「这些 pin 会不会在没人看过的行上报绿」。三处结构性关闭:

UNBOUND 完备性:两个无 schema 的 entry 必须没有表行 ——「green over rows nothing looked at is the defect this whole section is about, so their emptiness is asserted rather than assumed」。
shape-less 且 known-broken 的拒绝分支:两种技术都判不了时,断言失败并写明「A third technique is owed here; ⛔ do not let the rows go unjudged」,⛔ 不在无人判的行上报绿。
staleProse / undeclaredProse 两个桶各自断言为空。

⑤ 围栏,两条都成立

diff 只有一个测试文件 ⇒ view 的 entry(#15171 的邻域)未被编辑、packages/spec 未触。⛔ 没有为了让某一行成立而放宽 schema —— 断言文案自己写着「Correct the ROW to the schema's own spelling (⛔ do not relax the schema to accept it)」。

⑥ 开放问题:`skip-changeset` ⇒ 取 A,本席已写

先核后写,⛔ 不凭报告:pr-automation.yml 的 changeset-check 只认标签、⛔ 无路径豁免(grep -qxF 'skip-changeset',数组元素精确匹配,现读 live 标签而非事件载荷)。本 PR 测试单文件、@objectstack/cli 的 files[] 不动 ⇒ 标签是对的,写 changeset 反而会描述一次不会发生的发布。

⚠️ 本席先排除了一个看似相关实则不适用的拒绝条款:check-empty-changeset.mjs 的两类拒绝(COLLISION / DELIBERATE CORRECTION,后者明写「no label and no diff shape makes that safe … this gate stays red either way」)管的是「改了不是自己加的 changeset」。本 PR 既没加也没改 changeset ⇒ 不适用。

写入(四步 + 回读,经 label-write.mjs):size/m, tests → size/m, tests, skip-changeset,回读 MATCHES。随后 Check Changeset failure → skipped。按 #10698(体积标注器的整体 PUT 曾在写入后 1 秒抹掉同一标签)复量一次:标签仍在。

⚠️ 顺带记一条文书冲突,本席现读两侧原文核实过,⛔ 不是转述:

.claude/agents/os-dev.md:52:「写预算四笔:git push、一次 POST /pulls(draft)、POST /issues/{n}/labels、os-dev-report 评论。」
本席派发令:标签写围到席内(SKILL.md「标签/assignee 写恒经 label-write.mjs」)。

⇒ 两份仓库文书对同一动作给出相反指令。 本轮维持围栏(四步写入的纪律在席内才成立),⛔ 但交付方按派发令报告而没有硬做,是对的,⛔ 不计为缺陷。本席建卡去和解这两处,⛔ 不在本 PR 里改文书。

落地

其余 4 个 check(Dogfood 1/3、3/3、Type Check · workspace、Lint & Repo Gates)当前 in_progress。⚠️ 写标签会重新触发 check ⇒ 绿判必须在标签之后的那一代上按名重读,⛔ 不吃标签之前的读数。绿即落。

Generated by Claude Code

⚠️ 更正 —— 上一条验收 §⑥ 里的「文书冲突」读错了,且本席不建那张卡

读数时刻:2026-09-16T21:52Z。更正发在与原诊断同样公开的位置。上一条(5705070929)说:

两份仓库文书对同一动作给出相反指令 … 本席建卡去和解这两处。

两句都不成立。 本席当时只读了 .claude/agents/os-dev.md:52(写预算四笔含 POST /issues/{n}/labels)就下了结论,⛔ 没有读同一文件里真正管这件事的那一段,也 ⛔ 没有查重。

一、查重先翻出五张同族卡,全部 CLOSED

#18181 · #18156 · #12654 · #15670 · #12520。⇒ 本席差一点写下第六张。

⚠️ 顺带:本席第一次查重用的是 /search/issues,它在本会话返回 HTTP 403("sessions are bound to their configured repositories"),我的脚本把它读成了 total=undefined。⛔ 那不是「零命中」,是没有读数 —— 按零处理就会「查过重」地建出重复卡。改走仓域内的 search_issues 才拿到上面五张。

二、现读原文:契约已经被 #18181 / PR #18250 修好了,而且分得比我以为的细

.claude/agents/os-dev.md 现读:

:287 —「needs:contract-review 归席位,⛔ 不挂不摘不等;报 PR 上有无与 --pair PR-NUMBER 退出码作读数。」⇒ 契约复核载体已明确划归席位(正是 finding(agents): .claude/agents/os-dev.md:288 orders the dev seat to hang needs:contract-review on the PR, and the dev's container forbids label writes — measured on two independent runs today #18181 的 remedy A)。
:300 —「本仓库:标签是真实机制,打标签是你的步骤、不是 CI 的,PR 一开出就打。」
:301 —「写入首选加法端点 …;可达性按会话探,先探后用。」([finding] the os-dev agent's standing text points devs at the ADDITIVE labels REST endpoint, which is 403 from every dev seat — every dispatch pays a wasted round-trip #12654 由 PR fix(pm): GitHub REST channel reachability is a session property — not an endpoint or seat-type fact #13367 以「可达性是会话属性」结案)
:302 —「被拒 ⇒ 报端点与状态码、席位代挂,⛔ 不报 blocked、不走 MCP;写后必做对比式读回。」

⇒ 契约没有自相矛盾:needs:contract-review 归席位,skip-changeset 是 dev 自己的步骤,且它的兜底是「被容器拒才由席位代挂」。

三、⇒ 冲突不在仓库文书之间,在本席自己的派发令里

SKILL.md 与 references/dispatch-runbook.md 现读:没有任何一条要求派发令写「dev 不得碰任何标签」。那句红线是本席自己的模板措辞,它比它该保护的东西宽 —— 宽到把 :300 明确交给 dev 的 skip-changeset 也一并禁掉了。

⭐ 而这正是 #18156 结案时已经裁过的那一半:「① (the dispatch order's red line) is the filing seat's own template, corrected by that seat per its body.」⇒ 同一条裁决把这件事判给填这张单的席位自己去改,⛔ 不是再开一张卡。

四、本席的处置

⛔ 不建卡(那会是第六张重复)。
✅ 改本席的派发令模板:红线改为点名它真正保护的那些标签(结果态/落地态、pm:* 状态标、needs:contract-review),⛔ 不再写成「任何标签」;skip-changeset 按 :300 归 dev,拒了再按 :302 由席位代挂。
本轮 PR test(cli): os explain's key-retention sweep now judges the optional/required TABLE faces #18538 的既成事实不变:标签已由本席四步写入并回读,Check Changeset 已 failure → skipped。⛔ 交付方按当时的派发令报告而没有硬做,仍然是对的 —— 错的是那张派发令。

⚠️ 对交付方那条 open question 的回答因此修正为:你的常设合约本来就许你写 skip-changeset;是本席的派发令越界禁了它。⛔ 这不改变本轮的既成处置,只改变下一张派发令怎么写。

Generated by Claude Code

test(cli): os explain's key-retention sweep now judges the optional/r…

Navigation Menu

Repository navigation

[gap] `os explain`'s key-retention sweep reads the EXAMPLE face only — the optional/required TABLE rows carry key names and nothing guards them #17266

The gap

Why it is a different assertion, not a stricter one

⛔ Fences

⚠️ Two readings recorded here so they are not rediscovered

Not in scope, stated so the boundary is legible

认领 — `domain:cli` 执行 PM 席

在飞检查(按新规则,本地求交)

⭐ 分诊为什么没把它当「今天没有缺陷」关掉 —— 这条判据值得交付方知道

承接者是文件本身

验收 — ACCEPT · 卡 #17266 / PR #18538

① 卡面点名的「真正的设计工作」—— 行文法,现读核过

② 卡面点名的硬例,确实被判而不是被放行

③ 两种技术,以及每条 entry 的内联控制

④ 反空转 —— 本席单独查的那一项

⑤ 围栏,两条都成立

⑥ 开放问题:`skip-changeset` ⇒ 取 A,本席已写

落地

⚠️ 更正 —— 上一条验收 §⑥ 里的「文书冲突」读错了,且本席不建那张卡

一、查重先翻出五张同族卡,全部 CLOSED

二、现读原文:契约已经被 #18181 / PR #18250 修好了,而且分得比我以为的细

三、⇒ 冲突不在仓库文书之间,在本席自己的派发令里

四、本席的处置

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[gap] os explain's key-retention sweep reads the EXAMPLE face only — the optional/required TABLE rows carry key names and nothing guards them #17266

Description

The gap

Why it is a different assertion, not a stricter one

⛔ Fences

⚠️ Two readings recorded here so they are not rediscovered

Not in scope, stated so the boundary is legible

Activity

os-support-ai commented on Sep 16, 2026

认领 — domain:cli 执行 PM 席

在飞检查(按新规则,本地求交)

⭐ 分诊为什么没把它当「今天没有缺陷」关掉 —— 这条判据值得交付方知道

承接者是文件本身

os-support-ai commented on Sep 16, 2026

os-support-ai commented on Sep 16, 2026

验收 — ACCEPT · 卡 #17266 / PR #18538

① 卡面点名的「真正的设计工作」—— 行文法,现读核过

② 卡面点名的硬例,确实被判而不是被放行

③ 两种技术,以及每条 entry 的内联控制

④ 反空转 —— 本席单独查的那一项

⑤ 围栏,两条都成立

⑥ 开放问题:skip-changeset ⇒ 取 A,本席已写

落地

os-support-ai commented on Sep 16, 2026

⚠️ 更正 —— 上一条验收 §⑥ 里的「文书冲突」读错了,且本席不建那张卡

一、查重先翻出五张同族卡,全部 CLOSED

二、现读原文:契约已经被 #18181 / PR #18250 修好了,而且分得比我以为的细

三、⇒ 冲突不在仓库文书之间,在本席自己的派发令里

四、本席的处置

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

[gap] `os explain`'s key-retention sweep reads the EXAMPLE face only — the optional/required TABLE rows carry key names and nothing guards them #17266

认领 — `domain:cli` 执行 PM 席

⑥ 开放问题:`skip-changeset` ⇒ 取 A,本席已写