finding(docs/gate): check-doc-component-types 的扫描面止于代码围栏 —— 八个 PR 新加的「真实 register 键表」全部在门外,且门从不判 namespace 半边

发现于 #5094 的实施(PR #5104)反向验证。Filed unassigned, not claiming。观察类:今天没有读者会撞到 —— 现存的每张键表实测都是对的;这条记的是「守它的东西不存在」。

两条实测事实

scripts/check-doc-component-types.mjs 的 scanDocs 只在 fenced code block 内取 type: 'VALUE' 形的字面量(函数自己的注释写明:"Fences are tracked so prose that merely mentions a type in backticks is not read as a snippet")。由此:

1. markdown 表格 / 散文里的组件键,门看不见。

在 content/docs/plugins/plugin-dashboard.mdx 的键表里插一行假键:

| plugin-dashboard:phantom-widget | phantom-widget | PhantomWidget — nothing registers this |

node scripts/check-doc-component-types.mjs → rc=0,Every documented component type is registered.

同一个假键放进本页某个 fenced code block("type": "phantom-widget")则立刻红:

content/docs/plugins/plugin-dashboard.mdx:179  [unregistered-doc-type]  type 'phantom-widget' (json)

门本身有牙,牙只长在围栏里。

2. 门从不判 namespace 半边。

把 packages/plugin-dashboard/src/index.tsx 里 dashboard 那次注册的 namespace: 'view' 改成 'dash',deriveRegistryKeys 立刻翻成 dash:dashboard(推导是活读),但 check-doc-component-types 仍然全绿 —— 它判的是 bare 键存在与否。所以任何文档把 view:dashboard 写成 plugin-dashboard:dashboard,静态面零信号。

为什么现在值得记一笔

#5002 家族这一轮(PR #5071 / #5078 / #5079 / #5085 / #5089 / #5093 / #5100 / #5104)把「虚构的 *Components 手动注册循环」一律换成了同一种同族形:一张 | Namespaced key | Bare-name fallback | Renderer | 的 markdown 表。这个形是对的 —— 但它恰好整体落在门的扫描面之外,而它替换掉的那些代码块曾经在扫描面之内(尽管那批缺陷门也抓不到,原因另有其一:旧循环里 type 是解构出的标识符,不是 type: '字面量')。

净效果:这一族修复把「教什么键」的事实从代码块搬进了表格,守它的只剩人工比对。#4823(门的立卡)在正文里把扫描面定义成代码块,是有意的作用域,不是 bug —— 所以这条不是「门坏了」,是「同族形与门的作用域错开了」。

可能的方向(不预断,交分诊)

A. 把 | \ns:key` | `key` |这种**结构化表格行**也纳入扫描(只认这一种形,避免把散文里的反引号全部拖进来),namespaced 半边对deriveRegistryKeys` 的 namespace 一起核;
B. 不动门,改让这族键表生成而不是手抄(键表由 deriveRegistryKeys 产出,文档只留一处 include);
C. 判定为可接受的作用域,只在门的头注释里写明「表格不在扫描面」,让下一个照抄同族形的 agent 知道自己没有网。

代价与长期取向差别很大(A 扩门面、可能引入误红;B 是契约先行但要新增生成步骤;C 零成本但把可防错性留给人),故不自裁。

复核方式

sed -n '685,730p' scripts/check-doc-component-types.mjs      # scanDocs 的围栏跟踪
node scripts/check-doc-component-types.mjs                    # 基线绿
# 在 content/docs/plugins/plugin-dashboard.mdx 键表插一行假键 → 仍绿
# 把同一假键放进任一 json 代码块的 "type" → 红 unregistered-doc-type

同一扇门的第三条扫描面事实,发现于 #5115 的实施(PR 见下)。不另立卡,挂在这里。

本卡已记的两条是:(1) 围栏外的表格/散文看不见;(2) 门不判 namespace 半边。两条都是已扫描文件内部的取值范围问题。这条不同 —— 是 walkFiles 的文件过滤本身:

scripts/check-doc-component-types.mjs:694
  const files = walkFiles(docsDir, (f) => f.endsWith('.mdx')).sort();

content/docs 下有 143 个 .mdx 和 40 个 .md。后者整包在门外 —— 连它们的 fenced code block 也不扫,所以本卡事实 (1) 覆盖不到:那条讲的是被扫文件里围栏之外的内容,这条讲的是压根没被走到的文件。

实测

用本卡同一份 deriveRegistryKeys 推导跑那 40 个 .md 的围栏内 type 字面量:

off-registry type literals in content/docs/**.md: 31 distinct values, 67 sites
   10x  action          9x  string         5x  item          4x  board
    4x  array           3x  ObjectGrid     3x  my-component  2x  link
    2x  bar             2x  line           2x  theme         2x  empty-state
    1x  crud            1x  enum           1x  picklist      1x  my-grid   ...

多数是 #4823 头注释里列过的其它词汇表(string / array / enum 是 JSON-Schema,action 是 ActionSchema),放进门里需要同一套 exemption 才不误红 —— 这正是 A 方向的成本。但里面确实混着真幽灵:

content/docs/api/schema-reference.md:530 教 "type": "crud" —— 全仓零注册,渲染落 OBJUI-001 占位符(objectui check 的 knownTypes 是手写的注册表副本,已与真实注册表漂移:crud 通过校验但无任何渲染器,运行时落 unknown-component 占位符 #5115 实测两半);
content/docs/guide/building-crud-app.md:187 教 ObjectGrid(大小写错;真身 object-grid);
content/docs/guide/architecture.md:355 的 my-grid、component-registry.md:64 的 my-component 看形是占位示例,但门无从区分。

对本卡分级的一点提请

本卡自评「观察类:今天没有读者会撞到」。就已记的两条而言成立。加上这条之后不完全成立 —— schema-reference.md:530 是照抄即红的公开 API 文档页,与 #4786 / #4796 同形(那两条都是人肉探针发现的,不是门)。是否因此改级交分诊,我不自裁。

方向上这条与本卡的 A/B/C 三选并不冲突:A 若落地需要连 .md 一起走(且带上其它词汇表的 exemption),B/C 则对这 40 个文件无效。

复核:

find content/docs -name '*.md'  | wc -l    # 40
find content/docs -name '*.mdx' | wc -l    # 143
sed -n '694p' scripts/check-doc-component-types.mjs
grep -n '"type": "crud"' content/docs/api/schema-reference.md   # :530

Generated by Claude Code

Triage (first-touch, triage seat, session session_01HXzdkKx5WjwwCTT3c7U5kB): promoted finding → pm:queue, + tooling, type Task. Route decision is a verification-strategy call, so it is made here rather than escalated: route A adopted as scope, extended by this card's third comment.

Scope: (1) teach check-doc-component-types the structured key-table row form (| \ns:key` | `key` | … |only — do not drag prose backticks into scan), checking the namespaced half againstderiveRegistryKeys; (2) include the 40 content/docs/**/*.mdfiles inwalkFiles(today.mdx-only), which requires an exemption set for the other vocabularies measured in that comment (JSON-Schema string/array/enum, ActionSchema action, placeholder ids); (3) fix the two real ghosts that measurement surfaced, as this PR's validation corpus: content/docs/api/schema-reference.md:530teaches"type": "crud"(zero registrations, renders the OBJUI-001 placeholder — copy-it-and-it's-red on a public API doc page) andcontent/docs/guide/building-crud-app.md:187 ObjectGridcasing (真身object-grid`). Route B (generated key tables) recorded as a future candidate, not this card; C rejected — it leaves the family's new canonical form netless.

Size/model suggestion: M, mode:subagent.

Generated by Claude Code

PM claim + dispatch order — R27

Claimed by the domain:devx @ objectui execution seat (#5748), PM session session_019b5UBNMtTzKbVtZZGvFuxe. Branch: claude/issue-5106-doc-type-scan-surface.

Route A is adopted by triage — ⛔ not mine to re-open, and the scope is three parts

⚠️ The card body says 「故不自裁」 and leaves A/B/C open. That is superseded: triage adopted route A (a verification-strategy call it made rather than escalated), extended by this card's own third comment. ⛔ Route C was rejected — it leaves the family's new canonical form netless. Route B (generated key tables) is recorded as a future candidate, not this card.

Teach the gate the structured key-table row form — | \ns:key` | `key` | … |**only**. ⛔ Do not drag prose backticks into the scan; that is the false-positive surface the card warns about. Check the **namespaced** half againstderiveRegistryKeys`, which today is never judged at all.
Include the 40 content/docs/**/*.md files in walkFiles (today .mdx-only, at :694). ⚠️ This requires an exemption set for the other vocabularies the third comment measured — JSON-Schema string / array / enum, ActionSchema action, placeholder ids. Without it the gate reds on 31 distinct values across 67 sites that are not component keys at all.
Fix the two real ghosts that measurement surfaced — they are this PR's validation corpus, not a rider:
- content/docs/api/schema-reference.md:530 teaches "type": "crud" — zero registrations, renders the OBJUI-001 placeholder. Copy-it-and-it-is-red, on a public API doc page.
- content/docs/guide/building-crud-app.md:187 teaches ObjectGrid — casing wrong, the real key is object-grid.

⭐ The controls, taken from the card's own reproduction

The card already ran the discriminating experiment, so reproduce it as the before/after pair rather than inventing one:

A fake key in the markdown table (| plugin-dashboard:phantom-widget | … |) → gate green today. After your change it must red.
The same fake key in a fenced block → already reds today, and must still red. ⛔ Do not trade one surface for the other.
⭐ The namespace half: change a registration's namespace: 'view' to 'dash' and confirm the gate now notices. Today deriveRegistryKeys flips to dash:dashboard while the gate stays green, because it only judges the bare key. That is the second half of part 1 and it is easy to leave unimplemented while the table-row parsing looks done.
Exemption non-vacuity: the exemption set must not be so wide it swallows real ghosts. Prove it by confirming "type": "crud" at schema-reference.md:530 reds — a .md file, an unregistered key, and exactly the case the exemptions must not cover.

⛔ Mutations proven on disk by anchored counts, never an editor's exit code; restore via git checkout HEAD -- plus the explicit path, proven with an empty git diff HEAD.

Non-vacuity and scope

Population must refuse to collapse: zero files walked, zero key tables found, or zero registrations derived must fail, not pass. Put the census in the verdict line (.mdx count, .md count, table rows judged, fenced literals judged). scripts/check-vi-mock-specifiers.mjs (Nothing catches a vi.mock() whose relative specifier resolves to no file — vitest no-ops it and the suite stays green #5646, merged tonight) is the shape.
⚠️ Report the .md / .mdx counts you measure. The card says 143 / 40 as of 2026-08-18; re-derive — every inherited number tonight has moved.
⛔ Do not implement route B. ⛔ Do not fix any other ghost the widened scan surfaces — report it; the two named above are the corpus.

Root vitest only (objectui#3378). Exit codes captured by redirect before any pipe. ⚠️ Derive the gate list from package.json and .github/workflows/ rather than from this order.

⭐ When you edit or widen a gate, derive which tests read it and run them. A tooling diff is not automatically a tooling-only diff — that gap reddened PR #6189 tonight.

Generated by Claude Code

feat(tooling): read plugin key tables in the doc component-type gate,…

os-dev-report

{
  "issue": 5106,
  "status": "done",
  "branch": "claude/issue-5106-doc-type-scan-surface",
  "pr": "https://github.com/objectstack-ai/objectui/pull/6205",
  "premise_still_valid": false,
  "summary": "Part 1 of the order was live and is implemented: check-doc-component-types.mjs now reads plugin KEY TABLES and judges BOTH halves of every row, including the namespaced half it never judged at all. Parts 2 and 3 were already landed and are reported, not re-done — part 2 (.md inclusion) shipped as sub-issue #5342 (closed 2026-08-20; DOC_EXTENSIONS is already ['.mdx','.md'] with the ledger keyed by .md paths), and BOTH named ghosts are gone: `crud` was retired with CRUDSchema in #6117 (ADR-0049) and schema-reference.md now documents its removal, while building-crud-app.md's `ObjectGrid` is fixed AND already pinned by a test at check-doc-component-types.test.ts:458. The order's intended validation corpus therefore no longer exists, so the controls it specified serve as the corpus. KEY DESIGN CALL: the anchor is the table HEADER (`| Namespaced key | Bare-name fallback | … |`), not the row shape — the row heuristic was built and measured over this tree first and matches 33 rows of which only 22 are keys; the other 11 are React route patterns, URLs, HTTP routes and JSON literals, every one a false RED on correct docs. Separately found and fixed a dead floor inside the mechanism being extended: FLOORS.docFiles named a counter that never existed (scanDocs publishes `files`), so it compared undefined and the check meant to catch the docs walk finding NOTHING was inert for its whole life; the class is closed by a guard that fails loudly on any floor naming no counter.",
  "tests": "Union re-run on final commit 93124e4d5, all under scripts/pm/os-verify-lock.sh, exit codes captured by redirect before any pipe, verdicts quoted from each gate's own line. GATE: `node scripts/check-doc-component-types.mjs` -> DT_EXIT=0, verdict line 'Scanned 184 doc file(s) (.mdx + .md), 1057 code block(s), 895 `type` literal(s) against 659 registered key(s) derived from 136 source file(s) (235 resolved call site(s), 133 indirect, 2 open): 760 registered, 135 exempted; 4 key table(s), 24 row(s), 45 table key(s) judged (namespaced + bare), 45 registered.' RE-DERIVED COUNTS (card's 143/40 is from 2026-08-18): 144 .mdx + 40 .md = 184; 4 key tables, 24 rows, 45 judged keys, all registered. CONTROLS — each run against HEAD's gate AND this branch's gate on the IDENTICAL mutated tree (HEAD's copy extracted via `git show HEAD:...`), so the pair isolates the change, not the tree. Every mutation proven on disk by anchored counts; every restore proven by an empty `git diff HEAD`. (1) Fake key in the MARKDOWN TABLE at plugin-dashboard.mdx:107 (phantom count 0->1): HEAD's gate exit 0 'Every documented component type is registered.'; this branch exit 1 with 2 findings 'unregistered-key-table-key namespaced plugin-dashboard:phantom-widget' + 'bare fallback phantom-widget'. Green->RED as predicted. (2) Same fake key in a FENCED block at :155 (line text verified changed): HEAD exit 1, branch exit 1, IDENTICAL finding 'unregistered-doc-type type phantom-widget (json)' — no surface traded. HONEST NOTE: the first attempt at this control was a NO-OP — the perl -i anchor matched zero lines, exited 0, and the gate ran green, which would have read as 'fenced surface still works' while testing nothing; caught because the anchored count did not move (phantom stayed 0, actionLit unchanged). Re-run against a verified anchor; the reading above is the real one. (3) NAMESPACE HALF (the star control): plugin-dashboard/src/index.tsx:83 `namespace: 'view'` -> `'dash'` (view 1->0, dash 0->1, line text verified): HEAD's gate exit 0 FULLY GREEN — the card's finding reproduced exactly; this branch exit 1 'content/docs/plugins/plugin-dashboard.mdx:99 [unregistered-key-table-key] namespaced view:dashboard'. Exactly ONE finding, because the bare `dashboard` still matches — which is precisely why a bare-key-only gate can never notice a namespace move. (4) EXEMPTION NON-VACUITY: injected unregistered `\"type\": \"crud\"` into content/docs/api/schema-reference.md (crudLit 0->1, actionLit 5->4 proving the substitution) — a .md file that HAS exemptions (action, string): exit 1 'schema-reference.md:562 [unregistered-doc-type] type crud (json)'. Exemptions are keyed (file,value) and do not swallow it. NON-VACUITY: renaming the key-table header in all 4 files (occurrences 4->0, proven, then restored to 4) yields 'The scan collapsed: keyTables = 0, below the floor of 2.' — fails, does not pass. DEAD-FLOOR PROOF: re-introducing the exact original spelling `docFiles: 100` yields exit 1 'FLOORS names `docFiles`, which is not a counter this scan publishes (callSites, codeBlocks, exempted, files, indirect, keyTableKeys, ...). A floor over a missing counter compares `undefined` and can never fail, so it guards nothing.' TESTS (root vitest only, objectui#3378) — a tooling diff is not automatically tooling-only, so consumers were DERIVED by grepping importers: consumers import only deriveRegistryKeys (untouched); scanDocs/analyze changed additively. check-doc-component-types.test.ts: 49 passed (+12 new). Consumer set known-schema-types-derivation-5115 + check-pre-install-import-graph + check-doc-snippet-types + scripts-type-check + examples/schema-catalog/catalog-gallery-render: 5 files, 652 passed. Final-commit re-run of gate + 4 consumer files: 107 passed. OTHER GATES (derived from package.json scripts + .github/workflows/, not from the dispatch): check-control-bytes OK (5122 tracked text files) · type-check:scripts exit 0 · lint:coverage exit 0 · type-check:coverage exit 0 · check-changeset-presence exit 0 ('1 changeset(s) added') · check-pre-install-import-graph exit 0 · check-doc-fences exit 0 · control-byte self-scan of all 4 changed files: no matches. DECLARED NARROWINGS (CI runs the full farm regardless): eslint narrowed to the 2 changed source files — three-part proof: (a) population read from eslint's own config, which resolved and linted both, (b) file count 2 read from --format json with 0 errors / 0 warnings, (c) eslint.config.js enables no type-aware linting (no projectService, no parserOptions.project), so this diff cannot move any untouched file's verdict. check:doc-snippets DECLINED TO RUN — its own message: 'The snippet program was NOT run: the packages it resolves against are not built' — an unbuilt worktree, not a finding; this diff contains zero content/docs bytes.",
  "open_questions": [],
  "out_of_scope_findings": [
    "NOT filed as a new issue — fixed in-place under the bounded-exemption rule and named in the PR body with its proof: FLOORS.docFiles in scripts/check-doc-component-types.mjs named a counter that never existed, so the floor guarding against a collapsed docs walk compared `undefined` and could never fire. Same defect class as this card (the gate's own non-vacuity mechanism, which this PR extends), mechanical, evidence-pinned, same gate family, no new validation surface. The class is closed by a guard, not just the one instance.",
    "No new ghosts surfaced by the widened scan: all 45 key-table keys across the 4 tables are registered, so nothing was left unfixed-and-unreported. The card's 'my-grid' (guide/architecture.md) and 'my-component' (guide/component-registry.md) placeholder observations from the third comment remain covered by DOC_TYPE_EXEMPTIONS on the already-landed .md surface, not by this change.",
    "Reported for PM triage, not acted on: this card's own third comment asked whether the card should be re-graded from 'observation-class' now that schema-reference.md:530 was a copy-it-and-it-is-red public API page. That specimen has since been retired independently (#6117), so the re-grade question is moot and the card can close on this PR's part-1 work."
  ]
}

Generated by Claude Code

PM review — ACCEPTED. ⛔ Arming held (16 green, 8 running, 0 failures). ⚠️ Two thirds of my dispatch order were already done, and that is my error.

Reviewed against a pinned ref by the domain:devx @ objectui execution seat (#5748), PM session session_019b5UBNMtTzKbVtZZGvFuxe.

⛔ My order was stale, and I verified it myself rather than taking the report's word

I wrote a three-part scope. Parts 2 and 3 had already landed. Checked on origin/main:

what my order demanded	actual state on `main`
"include the 40 `content/docs/*/.md` files (today `.mdx`-only)"	`DOC_EXTENSIONS = ['.mdx', '.md']` at line 167 — already inclusive, shipped as sub-issue #5342, closed 2026-08-20
"fix the ghost `\"type\": \"crud\"` at `schema-reference.md:530`"	0 occurrences — `crud` was retired with `CRUDSchema` in #6117 (ADR-0049), and the page now documents its removal
"fix `ObjectGrid` casing at `building-crud-app.md:187`"	0 occurrences — already fixed, and already pinned by a test at `check-doc-component-types.test.ts:458`

So the validation corpus I specified no longer exists. The dev implemented part 1, reported the rest, and ⭐ did not invent work to fill the gap — the controls became the corpus instead. That is the right response to a stale order.

⚠️ The sharper version of my mistake: my own order said "⚠️ RE-DERIVE the 143/40 file counts — the card is from 2026-08-18 and every inherited number tonight has moved." I applied that reasoning to the numbers and not to the work items, which came from a triage comment of the same vintage. Staleness does not stop at figures. A dispatch order's scope needs re-deriving exactly as much as its counts do, and I have now been corrected on this class four times tonight.

⭐ The design call is the best thing in this PR, and it was measured rather than argued

My order said: match | \ns:key` | `key` | … |` only, and do not drag prose backticks in. They built that row heuristic, measured it over the real tree first, and rejected it:

the row heuristic matches 33 rows of which only 22 are keys — the other 11 are React route patterns, URLs, HTTP routes and JSON literals, every one a false RED on correct docs.

So the anchor became the table header (| Namespaced key | Bare-name fallback | … |) rather than the row shape. ⭐ That is exactly the false-positive surface I warned about, quantified instead of reasoned about — and had they followed my order literally, the gate would have reddened eleven correct documents on arrival and been deleted by the first person it annoyed.

⭐⭐ The dead floor — a non-vacuity mechanism that was itself vacuous

Found inside the very mechanism being extended, and I verified it on main:

FLOORS = { docFiles: 100 }     // line 505-506
… but scanDocs publishes `files`, not `docFiles`

counters['docFiles'] is undefined, so the comparison could never be true. The check meant to catch the docs walk finding nothing was inert for its entire life.

And they closed the class, not the instance — a guard that fails loudly on any floor naming a counter the scan does not publish, with the reasoning written at the site:

a collapse detector that can itself collapse is worth less than none, so this is checked before the comparison rather than left to review.

⭐ That is this lane's recurring subject turning up inside its own instrument. Proven by re-introducing the original spelling and observing the new error name the missing counter and list the real ones.

The controls are methodologically better than what I specified

⭐ Each was run against HEAD's gate and the branch's gate on the identical mutated tree (HEAD's copy extracted via git show), "so the pair isolates the change, not the tree." I asked for before/after; they gave same-tree/two-gates, which removes the confound I left open.

Table row: HEAD exit 0 "Every documented component type is registered" → branch exit 1 with two findings (namespaced + bare fallback). Green → red.
Fenced block: HEAD exit 1, branch exit 1, identical finding — no surface traded.
⭐ Namespace half — the star: flipping namespace: 'view' → 'dash' leaves HEAD's gate fully green, reproducing the card's finding exactly, while the branch reports exactly one finding. And the reason is stated: "the bare dashboard still matches — which is precisely why a bare-key-only gate can never notice a namespace move."
Exemption non-vacuity: an unregistered crud injected into a .md file that has exemptions still reds, because exemptions are keyed (file, value).
Collapse: renaming the table header in all four files yields "The scan collapsed: keyTables = 0, below the floor of 2" — fails, does not pass.

⭐ Third no-op caught tonight, and reported rather than buried

the first attempt at this control was a NO-OP — the perl -i anchor matched zero lines, exited 0, and the gate ran green, which would have read as "fenced surface still works" while testing nothing; caught because the anchored count did not move.

Three separate devs have now hit that trap in one evening and all three caught it the same way. The anchored-count discipline is earning its keep.

Landing

⛔ Not arming: Type Check, Lint, Doc Snippet Type Check, Live E2E and all four shards are still running. I will arm when every check concludes.

The card's third comment asked whether it should be re-graded from observation-class now that schema-reference.md was a copy-it-and-it-is-red public page. ⭐ That specimen was retired independently by #6117, so the question is moot and the card can close on this PR's part-1 work.

check:doc-snippets declined to run on an unbuilt worktree — correctly declared as a precondition, not a finding, and the diff contains zero content/docs bytes. ⚠️ That is the same false-green now dispatched as #5465's separable half.

Generated by Claude Code

H8 配对写入：摘 `pm:dispatched`，本卡关闭（completed）

交付 PR #6205 已于 2026-08-25T02:04:47Z 合并，本卡仍挂 pm:dispatched，已挂 5.5 天。

按 PM review（5403887696）末段执行，原文：

The card's third comment asked whether it should be re-graded from observation-class… That specimen was retired independently by #6117, so the question is moot and the card can close on this PR's part-1 work。

即：本卡原定三部分里，第 2、3 部分在派发前就已由 #5342 与 #6117 落地（DOC_EXTENSIONS 早已含 .md；crud 与 ObjectGrid 两处均 0 命中且后者已被测试钉住），第 1 部分随本 PR 交付，验证语料改由控制组承担。

⭐ 随卡留存一条值得记的发现：FLOORS = { docFiles: 100 } 而 scanDocs 发布的是 files —— 那个防塌陷检查终其一生都是惰性的，本 PR 连类一起关掉了（floor 命名了扫描未发布的计数器即报错）。

本笔是维护者指令下的 H8 清账（2026-08-30，session session_014kugUSM5M5fJBsk1f8KdtN）：按该席位已记录的处置执行，⛔ 未改卡面范围、未动 assignee。

Generated by Claude Code

feat(tooling): read plugin key tables in the doc component-type gate,…

Navigation Menu

Repository navigation

finding(docs/gate): check-doc-component-types 的扫描面止于代码围栏 —— 八个 PR 新加的「真实 register 键表」全部在门外,且门从不判 namespace 半边 #5106

两条实测事实

为什么现在值得记一笔

可能的方向(不预断,交分诊)

复核方式

实测

对本卡分级的一点提请

PM claim + dispatch order — R27

Route A is adopted by triage — ⛔ not mine to re-open, and the scope is three parts

⭐ The controls, taken from the card's own reproduction

Non-vacuity and scope

PM review — ACCEPTED. ⛔ Arming held (16 green, 8 running, 0 failures). ⚠️ Two thirds of my dispatch order were already done, and that is my error.

⛔ My order was stale, and I verified it myself rather than taking the report's word

⭐ The design call is the best thing in this PR, and it was measured rather than argued

⭐⭐ The dead floor — a non-vacuity mechanism that was itself vacuous

The controls are methodologically better than what I specified

⭐ Third no-op caught tonight, and reported rather than buried

Landing

H8 配对写入：摘 `pm:dispatched`，本卡关闭（completed）

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(docs/gate): check-doc-component-types 的扫描面止于代码围栏 —— 八个 PR 新加的「真实 register 键表」全部在门外,且门从不判 namespace 半边 #5106

Description

两条实测事实

为什么现在值得记一笔

可能的方向(不预断,交分诊)

复核方式

Activity

yinlianghui commented on Aug 18, 2026

实测

对本卡分级的一点提请

os-support-ai commented on Aug 19, 2026

yinlianghui-tw commented on Aug 25, 2026

PM claim + dispatch order — R27

Route A is adopted by triage — ⛔ not mine to re-open, and the scope is three parts

⭐ The controls, taken from the card's own reproduction

Non-vacuity and scope

yinlianghui-tw commented on Aug 25, 2026

yinlianghui-tw commented on Aug 25, 2026

PM review — ACCEPTED. ⛔ Arming held (16 green, 8 running, 0 failures). ⚠️ Two thirds of my dispatch order were already done, and that is my error.

⛔ My order was stale, and I verified it myself rather than taking the report's word

⭐ The design call is the best thing in this PR, and it was measured rather than argued

⭐⭐ The dead floor — a non-vacuity mechanism that was itself vacuous

The controls are methodologically better than what I specified

⭐ Third no-op caught tonight, and reported rather than buried

Landing

os-project-manager commented on Aug 30, 2026

H8 配对写入：摘 pm:dispatched，本卡关闭（completed）

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

H8 配对写入：摘 `pm:dispatched`，本卡关闭（completed）