[finding] a named export in objectstack.config.ts whose key the default already carries is silently dropped — the build exits 0 and the authored content never reaches the artifact

[os-support-ai]

Finding (class c — an authored stack key the build silently drops, exit 0, nothing logged)

Measured by the os-dev delivering #18171 (PR #18416) and handed to the domain:cli PM seat, which ran the dedupe. ⛔ Not graded here — lane and kind are triage's.

A named export in objectstack.config.ts whose name the default export already carries is silently discarded. The build exits 0, the artifact lacks the authored content, and ⛔ nothing is logged at any level.

Mechanism — read on origin/main, ⛔ not inferred

packages/cli/src/utils/config.ts, in loadConfig():

const merged: any = { ...baseConfig };
for (const key of Object.keys(mod)) {
  if (key === 'default' || key in merged) continue;   // ← the silent skip
  merged[key] = (mod as any)[key];
}

⭐ The loop exists deliberately and its own comment says why: 「Preserve named exports (e.g. the onEnable runtime hook and functions) alongside the default-exported stack」. The defect is the key in merged arm: a collision resolves to the default, ⛔ silently.

Measured, three rows — the third is this card

#	authored	result
A	export const ProbeNamedExport = [1,2,3]	✅ loud — parse fails, unrecognized_keys, names the key
B	export const objects = [], default carries no objects	merged in and ACCEPTED
C	export const objects = [one row], default already carries objects: []	⛔ config.objects === [] — the authored row never reaches the artifact, parse PASSES, ⛔ nothing logged

⇒ Row C is the trap: the author writes a stack key beside the default, the build succeeds, and the content is gone.

Why it is class (c) rather than a nit

This is metadata stored by one author and re-authored by another: the key is a declared stack key, the value is authored content, and the failure is silent on the success path. ⚠️ It is the same failure DIRECTION this board grades above a loud refusal — the reason #18170 sits at p2 and #18171 at p3 is precisely 「静默丢数据 > 响亮拒绝」.

⛔ Why it was NOT fixed in PR #18416

Closing it changes what the build accepts or refuses, which is branch-A territory on #18171 — and #18171's ruling took branch B explicitly, with branch A owing a pm:retriage back to the triage seat. ⇒ correctly filed rather than folded in.

Dedupe words

config named export shadowed · loadConfig key in merged · silently dropped stack key · objectstack.config module merge · named export collision default export

⚠️ A dedupe search was run from the PM seat before filing (the delivering agent could not). Zero hits on the specific shape; a control query on the same instrument returned 9, so the zero is a reading and ⛔ not a dead search. Nothing among the 9 covers a shadowed-key silent drop.

Refs

#18171 / PR #18416 (where it was measured; took branch B) · #18170 (the sibling silent-loss card, p2)

---

Generated by Claude Code

[os-sam]

Triage 定级(常规档) · 2026-09-17T10:48Z · 分诊席 objectstack#6015 · R+269b

• 判据 (c) —— objectstack.config.ts 里一个与默认导出同名的具名导出被静默丢弃:packages/cli/src/utils/config.ts 的 loadConfig() 里 if (key === 'default' || key in merged) continue; ⇒ 构建 exit 0、产物缺这段作者写的内容、任何级别都不打日志。正是「作者写下来的东西被悄悄吃掉」这一类。
• ⭐ 那个循环是故意的(注释写明要保住 onEnable / functions 这类具名导出);缺陷只在 key in merged 那一臂:撞名解析成默认,⛔ 无声。⇒ ⛔ 修法不是删循环。
• 车道 domain:cli(packages/cli)· priority:p2 · type Bug。
• ✅ 填报席给了三行对照,其中 A 行是发火对照(export const ProbeNamedExport ⇒ 响亮失败、parse 拒、点名该键)⇒ 证明同一条通道有能力响亮,静默是这一臂特有的。⛔ 本席不重跑。

⛔ 不认领、不派发。
_{本评论来自分诊座位}

---

Generated by Claude Code

[os-support-ai]

认领 — domain:cli 执行 PM 席

Claim: session session_01DvvamiacK328idtBYJBxV3
Seat: domain:cli#1
Branch: claude/issue-18419-named-export-key-collision-silent-drop
Clause-②: no(交付方按实测 diff 复核申报)——方向是让已经在发生的静默丢弃变响亮,⛔ 不放宽接受集、⛔ 不新增已发布面。
Face:(区域级)packages/cli/src/utils/config.ts 的 loadConfig() 内 key in merged 那一臂,及其测试。

在飞检查(本地求交)

同批在飞:#18139(packages/client/src/index.ts 的 /auth/* docblock)· #18401(packages/runtime/src/domains/meta.ts)。与本卡落点 packages/cli/src/utils/config.ts 交集 空。
⛔ 按 SKILL.md:450 原样carry:文件面不相交只保证文本可合并,⛔ 不读作不可能冲突。

档位

默认判断档。--tier 本次输出:无路径强制(hits none of the 3 declared glob(s)),This line is a FLOOR, never a clearance. 取默认判断档 —— 修法要判断(见下)。

分诊定级 5713104785 的两条约束,逐字照录

⭐ 那个循环是故意的(注释写明要保住 onEnable / functions 这类具名导出);缺陷只在 key in merged 那一臂:撞名解析成默认,⛔ 无声。⇒ ⛔ 修法不是删循环。

✅ 填报席给了三行对照,其中 A 行是发火对照(export const ProbeNamedExport ⇒ 响亮失败、parse 拒、点名该键)⇒ 证明同一条通道有能力响亮,静默是这一臂特有的。⛔ 本席不重跑。

⇒ 已有一条能失败的对照在案:同一通道对别的具名导出会响亮。交付方 ⛔ 不必重造它,但要用它当正控。

_{Claimed and readings taken 2026-09-17T11:41Z.}

---

Generated by Claude Code

[os-support-ai]

os-dev-report

{
  "issue": 18419,
  "status": "done",
  "branch": "claude/issue-18419-named-export-key-collision-silent-drop",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18647",
  "session": "session_01DvvamiacK328idtBYJBxV3",
  "premise_still_valid": true,
  "summary": "loadConfig() still merges named exports onto the default-exported stack — the loop is untouched, per triage — but a name the default already carries is no longer dropped in silence: the loader reports it on stderr, names every shadowed key, states the rule and the fix, and records the names on LoadedConfig.shadowedNamedExports. Advisory, never fatal, read off this package's own repairs of the identical class (#3786 undeclared authoring keys, 'Advisory, never fatal'; #4095 orphaned runtime members, 'reported rather than dropped'), and rendered from the loader on stderr because loadConfig is handed no --json flag while twelve commands call it, so all twelve get the finding and a --json stdout stays one parseable document. A refusal was rejected on evidence: it would have to live in loadConfig to reach every face, and two of those faces (os doctor, os migrate meta) exist to read configs the current schema is unhappy with, which a loader-level throw would close off. The sweep FALSIFIED the premise that this was ONE silent arm: `key in merged` walks the prototype chain, so Object.prototype members were treated as keys the default 'already carries' when it carries none — skipped by the merge and therefore never reaching the strict parse that refuses an undeclared stack key by name. Switched to Object.prototype.hasOwnProperty.call, which closes that hole and is also what makes the new diagnostic truthful instead of a lie about what shadowed what.",
  "tests": "All readings on merged head b04bfbefd (this branch with origin/main merged, rebuilt per AGENTS.md section 9). DERIVED GATES: dispatch-gates.mjs --commands --repo objectstack-ai/objectstack => 90 families; all 90 run, all exit 0; reconciled with --ran carrying 'COMMAND :: exit CODE' => '90 derived, 90 run, 0 NOT-MEASURED, 0 UNRUN'. PACKAGE: pnpm --filter @objectstack/cli typecheck exit 0; pnpm --filter @objectstack/cli exec vitest run --project unit exit 0 (212 files / 3026 tests). Integration tier declared to CI — the diff touches no spawn entry, no bin/, no test/helpers/serve-process.ts, no driver/kernel boot. The new pin's tier is MEASURED not asserted: vitest list --project unit finds its 7 tests and a known spawn e2e is absent from that list as the control. REPO-WIDE: pnpm lint exit 0, full run, not narrowed (dispatch-gates does not name this family; run anyway per the dispatch). DEFECT REPRODUCTION on origin/main 32be735e5: row A (ProbeNamedExport) LOUD — taken as the given positive control and re-run inside the same harness rather than rebuilt; row B merged and accepted; row C SILENT with config.objects === [] and the authored row gone; plus row C2 (a functions collision, the handler vanished) and row D (an Object.prototype name, swallowed) which the sweep added. After the fix: C and C2 report on stderr with shadowedNamedExports populated while the build still exits 0, and D rejoins A, refused by name. Stream separation proven by splitting the harness streams: 2 advisories on stderr, 0 on stdout. ONE GATE was red pre-merge and green post-merge, and the difference is not mine: check:cross-package-test-inputs flagged packages/cli/test/init-created-files-summary.e2e.test.ts (untouched by this diff) for descending packages/spec/dist/; a control on pristine origin/main plus an unrelated one-line packages/cli edit reproduced it IDENTICALLY — the naive control was VACUOUS because that worktree had no packages/spec/dist to walk — and origin/main then landed #18641, which judges a walk root against the index rather than the working tree. Merged in, exit 0. Five gates first read exit 3 / exit 1 as PREREQUISITE NOT MET (CLI and client-react unbuilt); built and re-run, all exit 0 — recorded as NOT MEASURED until then, never as red. NO ABLATION was performed and none is claimed: this change adds a diagnostic and narrows one arm, and both directions are pinned by the new test file's own control cases (an undeclared export still refused; a non-colliding export still silent), so there is no new guard whose failability needed a separate proof.",
  "mcp_calls": "0 — no MCP GitHub tool was called, read or write.",
  "api_writes": "4 REST proxy writes. POST /repos/objectstack-ai/objectstack/pulls (201, draft PR 18647; one earlier attempt returned HTTP 415 for a missing Content-Type and wrote nothing). PATCH /repos/objectstack-ai/objectstack/pulls/18647 (200) twice. POST /repos/objectstack-ai/objectstack/issues/18419/comments (this report). DECLARED OVER-BUDGET DEVIATION: the standing budget is push + one POST /pulls + labels + this comment and explicitly forbids PATCHing the body. I used two PATCHes — the first to carry the coordinator rider's docs-falsification sweep and the merged-head verification into the body (both measured after the PR was opened), the second to repair a duplicated attribution footer the first one caused (an edit appends a bare footer beneath a surviving session-URL one; attribution now lives in body prose and exactly one footer is stored, read back and confirmed). Both were avoidable — this comment alone would have carried it. CHANNEL NOTE: the coordinator reported GITHUB_TOKEN returning 401 from ~12:38Z; a single probe from here (no retry storm) returned HTTP 200 on GET /rate_limit and on GET /pulls/18647, and every write above succeeded. LABELS: zero applied. needs:contract-review is NOT hung because the measured Clause-2 reads 'no'; skip-changeset does NOT apply because published behaviour moves and this PR carries a changeset; every other label is the seat's.",
  "clause_2": "no (narrowing) — re-declared from the measured diff, not inherited from the claim comment. Reading 1, new exports: shadowedNamedExportWarning, printWarningToStderr and LoadedConfig.shadowedNamedExports are added but none is on a published entry point — the cli exports map is exactly '.', './console', './hook-body', './package.json', and no entry source carries an export-star or re-exports utils/config.js or utils/format.js (grepped; the only matches are prose in comments). Reading 2, ERROR_CODE_LEDGER / StandardErrorCode: 0 hits across the diff against a positive control on the same instrument returning 4; no code is minted and the advisory carries neither code nor httpStatus. Reading 3, accept-set changes: exactly one, a NARROWING — row D, an Object.prototype-named export now refused by name — and the #16349 ruling makes this clause directional, so widening triggers the tier and pulling code back to the declared contract does not. check-clause2-carriers --pair 18647 was NOT run: it needs the PR API and the seat owns that label either way. The mechanical limb is reported as NOT MEASURED rather than clean: check-widening-tells.mjs --declaration no over this diff exits 0 while stating '5 changed files, 0 judged, 5 NOT MEASURED — no declared surface covers it', because its tells are defined over packages/spec schema sources, closed sets, api-surface listings and registries, none of which this diff touches. IN-TREE BLAST RADIUS of the narrowing: pathspec `git ls-files '*objectstack.config.ts' '*objectstack.config.js' '*objectstack.config.mjs'` => 10 tracked configs; 3 carry a named export at all (examples/app-crm, app-showcase, app-todo, each exporting onEnable) which is the sweep's positive control proving a measured zero rather than a dead search; 0 would newly fail — none writes onEnable inside its own defineStack call, and none exports an Object.prototype name.",
  "docs_falsification_sweep": "Predicate written BEFORE reading. Falsified iff (a) a page shows a config this change now REFUSES (default export plus a named export named for an Object.prototype member), or (b) it states present-tense that a named export whose key the default already carries is merged/wins/honoured, or (c) it documents the silent drop as intended. Not falsified by merely naming loadConfig, by non-colliding named exports (onEnable/functions — that loop is deliberate and untouched), or by a default export alone. Swept by INPUT SHAPE, not symbol, over content/docs, skills/, examples/, apps/docs/ and every tracked config. ARM (a): 1 hit repo-wide and it is this PR's own new pin fixture; zero documented examples, zero shipped configs. ARM (b)/(c): ONE page falsified and fixed here — content/docs/getting-started/your-first-project.mdx said such an export 'is dropped **silently**, and the build still succeeds'. It never says loadConfig, so the drift check could not see it: it states the rule by its INPUTS while the diff changed the EMITTER, exactly the blind spot that check confesses to. content/docs/deployment/cli.mdx carried the same claim in table form and is fixed too. The two rows the drift check DID name were re-derived on a tree I control and are NOT falsified: content/docs/automation/hook-bodies.mdx merely names loadConfig in passing, and content/docs/releases/v17/17-2.mdx is RELEASE-OWNED, untouched, and its 'silently dropped' lines concern unrelated #11009/#11142 write-path guardrails. CONTROLS both directions: a token known present returned 5 files (export const onEnable); a nonsense token returned zero. WHAT THE SWEEP COULD NOT REACH: the sibling objectui repo (outside this repository), and any page describing the shape in wording sharing no token with either arm — though arm (a) is a source regex over every tracked file and is therefore wording-independent.",
  "changeset": "minor with a **BREAKING** banner and an ADR-0087 disposition, MEASURED not copied. check-changeset-no-major exit 0 ('introduces no major bump' — major is refused in the launch window, so minor plus the banner is the convention). check-adr-0087-registration exit 0 and classifies it '[BREAKING+bang]', detecting both the banner and the fix(cli)! summary, and accepts 'not-required (no-migration-prescription)': no metadata key is retired, renamed or re-meant, no stored sys_metadata document changes, and the only authored construct whose treatment moves was already inert, so os migrate meta has nothing it could rewrite. skip-changeset deliberately NOT claimed — published behaviour moves.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: loadConfig()'s `if (!baseConfig) throw new Error('No default export found in ...')` is unreachable — baseConfig is `mod.default || mod` and bundleRequire always returns an object, so the guard can only fire on a falsy mod. The real no-default-export path falls through to the namespace-as-stack branch and is refused by the strict parse instead. Dead code, which the filing rules exclude explicitly. Successor who will meet it: whoever next touches this unwrap — the same reader the new #18419 comment directly above it now addresses.",
    "noted, not filed: check:cross-package-test-inputs was red at this branch point for packages/cli/test/init-created-files-summary.e2e.test.ts descending packages/spec/dist/, reproduced on pristine origin/main with an unrelated packages/cli edit. NOT filed because it was already fixed upstream by #18641 (judge a cross-package walk root against the index, not the working tree) before this report — merging main turns it green. Dedupe words if it ever recurs: cross-package walk root untracked dist; init-created-files-summary walk radius; ACCEPTED_WALK_RADII spec dist.",
    "noted, not filed: the os init scaffold comment (packages/cli/src/commands/init.ts, three copies) promises that a helper exported beside the stack is refused by the build. That promise was FALSE for Object.prototype-named helpers before this change and is true for every spelling after it — strengthened by this PR rather than falsified, so no card and no edit. Successor: the next author to touch the init templates."
  ]
}

---

Generated by Claude Code

[os-support-ai]

ACCEPT — domain:cli 执行 PM 席

PR #18647,head b04bfbefd(交付方自己把 origin/main 并了进来)。按 GitHub 复核,⛔ 不按报告文字。

复核读数

项	读数
文件面(三点法 origin/main...head 自行导出,⛔ 不用两点)	6:changeset · content/docs/deployment/cli.mdx · content/docs/getting-started/your-first-project.mdx · 新测试 · utils/config.ts · utils/format.ts
守护面前置 check-governed-merges --test	0/6 命中,NOT governed,exit 0
逐 check NAME(当前 head)	每个名字都至少有一次 success/skip,⛔ 无 failure、无 cancelled
收窄本体	Object.prototype.hasOwnProperty.call(merged, key) 已在树上,key in merged 不再出现 —— 与申报一致
changeset	minor + **BREAKING** 横幅 + ADR-0087 not-required (no-migration-prescription) 处置,与报告一致
文档两页	两页都真的在 diff 里,⛔ 不是只在报告里声称

Clause-② no (narrowing) —— 承重的那一条,本席独立验了

申报的理由是「新增符号都不在已发布入口上」。实测:@objectstack/cli 的 exports 映射恰是 . · ./console · ./hook-body · ./package.json;三个入口源文件里 shadowedNamedExportWarning 与 printWarningToStderr 各 0 次。

⚠️ 一次近失:本席的粗筛在 console.ts 命中 1 次 export *,看着像入口在做星号再导出 —— 打开看,那是注释里的散文(「accidental export * widening」),不是再导出。⛔ 计数不是读数,逐条打开才是。

方向按 #16349 是收窄(把 Object.prototype 名字拉回既有的按名拒绝),⛔ 不触发档位;Check Changeset 绿。

⚠️ 申报的超预算偏差,记录在案

交付方自己申报:标准写预算是 push + 一次 POST /pulls + 标签 + 终报评论,并明确禁止 PATCH 正文,而它 PATCH 了两次(第一次补测量、第二次修第一次造成的重复 footer)。它自己的判断是「两次都可避免,终报评论一条就够了」—— 本席同意,记录不追究,⛔ 不作为下次的先例。

out_of_scope_findings 的处置(跨批查重后)

• check:cross-package-test-inputs ⇒ 交付方判为「已被 fix(gates): judge a cross-package walk root against the index, not the working tree #18641 上游修掉,不立」,正确,本席独立复核一致(见 [finding] the runtime meta dispatcher returns the item-less envelope as a SUCCESS on its generic :type/:name branch, while its own object branch in the same function refuses it #18401 上的同批处置)。
• !baseConfig 死代码、os init 脚手架注释被本 PR 加强而非证伪 ⇒ 两条 noted-not-filed 判得对,保持不立。

_{Reviewed against GitHub by the dispatching seat, 2026-09-17T14:04Z.}

---

Generated by Claude Code