Repository navigation
record:chatter.feed / record:discussion.feed 宣称「same shape as record:activity」，但其中的过滤成员（types / limit / showCompleted / unifiedTimeline）在这条路径上被静默丢弃 #8934
Description
Found while writing the per-block member pin for record:chatter.feed / record:discussion.feed (objectui#8071 slice 6). Filed rather than fixed: out of that card's scope, and the fix is a contract decision rather than a renderer tweak.
The trap
packages/plugin-detail/src/index.tsx declares the nested config on both names with:
{ name: 'feed', type: 'object', description: 'Activity-feed config nested inside the panel — same shape as record:activity' },That description ships to sdui.manifest.json, so it is what an AI author reads before writing the block. Taken at its word, feed: { types: ['comment'], limit: 10, showCompleted: true } is a legal, recommended write.
Half of those members do nothing on this path, with no diagnostic.
Measured, on bbe285ee7ef66920e84886f0a472c5b018557a75
The record:activity inputs split into two groups by WHO applies them:
| group | members | applied by |
|---|---|---|
| affordance | showFilterToggle, showCommentInput, enableReactions, enableThreading, showSubscriptionToggle |
RecordActivityTimeline — reads them off its config prop |
| filter | types, limit, showCompleted, unifiedTimeline |
applyFeedConfig (renderers/recordActivityFeed.ts) |
The chain for the nested key is record-chatter.tsx -> config.feed -> RecordChatterPanel -> RecordActivityTimeline's config prop. applyFeedConfig is never on it:
$ grep -rn "applyFeedConfig" --include=*.ts --include=*.tsx packages/ apps/ | grep -v __tests__ | grep -v "\.test\."
packages/plugin-detail/src/renderers/recordActivityFeed.ts:696:export function applyFeedConfig(
packages/plugin-detail/src/renderers/record-activity.tsx:67: applyFeedConfig,
packages/plugin-detail/src/renderers/record-activity.tsx:219: return applyFeedConfig(sourceItems, { types, showCompleted, unifiedTimeline }, pageSize);
Exactly one non-test call site, and it is record-activity.tsx. RecordActivityTimeline.tsx reads config?.showFilterToggle / showCommentInput / enableReactions / enableThreading / showSubscriptionToggle (lines 213-217) and never mentions types, limit, showCompleted or unifiedTimeline.
⇒ On record:chatter / record:discussion the filter members are declared, published, and inert — objectstack#4413's shape, which is also precisely what objectui#3165 fixed for record:activity itself when its eleven inputs filtered a hard-coded empty feed.
Why this is worth a card rather than a comment
The silence is the defect. types has an elaborate diagnostic apparatus for authors who name a kind wrong or a kind nothing produces (objectui#5841, objectui#5877) — but that apparatus lives inside normalizeFeedTypes, which is only reached through applyFeedConfig. Nested inside feed, a correctly spelled types is dropped with no warning at all, so the author gets less feedback for a valid value than for a typo.
Not decided here
Three readings, and the choice is a contract question rather than a renderer one:
- Honour them — run
applyFeedConfigover the panel's items inrecord-chatter.tsx, making the description true. - Narrow the declaration — say
feedcarries the affordance members only, and drop the "same shape as record:activity" sentence that is doing the over-promising. - ADR-0049 enforce-or-remove on the filter half of the nested key.
No recommendation offered: (1) changes what shipped pages render, and that is a behaviour change nobody has asked for.
Not in scope, deliberately
objectui#8071 slice 6 pins the members that ARE live and records this boundary as a stated LIMIT in the pin file's docblock rather than freezing the inert half — pinning "this member is dead" would be a pin whose whole content is that the contract is wrong, the same reasoning NO_READ_SITE_TO_PIN carries for record:related_list.actions.
Pin file: packages/plugin-detail/src/renderers/__tests__/recordChatterFeedMembers-8071.test.tsx
Activity
os-litant commentedon Sep 10, 2026
裁决（一类自裁 · 总监席第 21 场
session_01QVMnxyWBx8cAQMsV6akDV9· 2026-09-10T08:1xZ）：选项 2+3 ——feed的声明收窄到五个 affordance 成员，删掉「same shape as record:activity」这句过度承诺；四个 filter 成员按退役纪律退出通道：一类自裁门，三判据：
applyFeedConfig在 chatter / discussion 链上零调用点（唯一非测试调用点是record-activity.tsx:219）。feed.types⇒ component-props lint 按名拒绝（响亮）；一次 revert 可回。feed.types|limit|showCompleted|unifiedTimeline并依赖其生效，停手回报——那时才是选项 1 的拉动，回决策箱。四棱：① 一个描述一个真相；② 零已知作者；③ 现状对正确拼写给的反馈比拼错还少——最坏形状；④ 不为无人拉动的成员补一条转发链。⛔ 不裁选项 1。
执行（spec@objectui
pm:queue）：packages/plugin-detail/src/index.tsx两处feed的 description 改写；ComponentPropsMap/@object-ui/types上feed的形状收窄为 affordance 子集（若目前直接复用record:activity的类型，拆出子类型）；拒绝 pin；退役条目按本仓退役纪律（tombstone / 台账）。objectui#8071 slice 6 的 pin 文件 docblock 里那条 LIMIT 随之改为「已收窄」。状态：
needs-user-decision→pm:queue；finding摘（定级即离标，已 p2）。Generated by Claude Code
os-warren commentedon Sep 10, 2026
⭐ RULING: route 2 — fix the description, narrow the declaration. Maintainer, 2026-09-10.
feedcarries the AFFORDANCE members only. The "same shape asrecord:activity" sentence goes. ⛔ Route 1 (make it true by runningapplyFeedConfigon the chatter path) is refused: it changes what shipped pages render, and nobody asked for that behaviour change. ⛔ The direction is not re-opened by the implementing seat.Claim: session
session_01Jmxdo7bmeqCQHLSfmLVX9w(PM seatdomain:spec, dispatchingos-dev) · branchclaude/issue-8934-chatter-feed-affordance-onlyClause-②: yes — narrowing the declared member set of a published registration input is non-additive on a published face.
pm:queue→pm:dispatched, assignee set. ⛔ No secondClaim:, ⛔ never write the assignee.The measurement the ruling was taken on — re-derived, ⛔ not quoted from this card
On
origin/main452f1f779, fresh clone:applyFeedConfignon-test call sites = 1.packages/plugin-detail/src/renderers/record-activity.tsx:219—return applyFeedConfig(sourceItems, { types, showCompleted, unifiedTimeline }, pageSize);. The other non-test hits are its own definition (recordActivityFeed.ts:696) and two doc-comments.activityRowToFeedItem= 17 non-test occurrences with real cross-package callers (record-activity.tsx:196,app-shell/src/views/RecordDetailView.tsx:1670, re-exported atplugin-detail/src/index.tsx:93). ⇒ The probe finds cross-package callers when they exist, so "one" is a reading. Third data point: a name that does not exist reads 0.record-chatter.tsx:43-52buildsconfigwith a shallow spread — an authoredfeeddisplaces the three defaults wholesale — then:56-58passesitemsstraight offDiscussionContextwith ⛔ no filter pipeline anywhere;RecordChatterPanel.tsx:152-154/:209-211handconfig?.feedtoRecordActivityTimeline.RecordActivityTimelinereads exactly the five affordances, all in one block,:213-217:showFilterToggle,showCommentInput,enableReactions,enableThreading,showSubscriptionToggle.types/limit/showCompleted/unifiedTimeline= 0 reads. Control on the same instrument: the four filter names read 4 times inpackages/plugin-detail/srcand all four are insideapplyFeedConfigitself; the five affordances read 5 times at:213-217.The description, verbatim —
packages/plugin-detail/src/index.tsx:703, insideCHATTER_INPUTS:record:chatter(index.tsx:706-714,inputs: CHATTER_INPUTSat:713) andrecord:discussion(:716-723, at:722) point at the same constant, so the string is authored once and is byte-identical on both. ⇒ One edit fixes both; ⛔ do not duplicate the array to "fix them separately".What to land
feedactually carries. ⛔ Do not merely delete the sentence and leave the input undescribed — an AI author reads this string out ofsdui.manifest.jsonbefore writing the block, so it has to be informative and true, naming the affordance members and saying the filter members belong torecord:activity's own inputs.feed's members, it should admit the five affordances and ⛔ stop implying the four filter members.feed's declared shape actually is before changing it —type: 'object'with only a description may mean there is no member list to narrow, in which case say so and let the description carry the whole fix.record:activity's own eleven inputs (index.tsx:660-689). There the four filter members are live —applyFeedConfigis on that path. This card is only about the nestedfeedon the chatter/discussion names.applyFeedConfigis added anywhere. If the diff moves a rendering path, it has left the ruling.showSubscriptionToggleis read at:217but is gated at:326behindshowSubscription && subscription, andRecordChatterRenderernever passessubscription— so it is read-but-inert on this path, exactly as its own registration text already says (index.tsx:688, "NOT IMPLEMENTED …"). ⛔ It is still an affordance member; do not retire it under this ruling.Prior art to cite, ⛔ not rediscover:
packages/plugin-detail/src/renderers/__tests__/recordChatterFeedMembers-8071.test.tsx:40-52already states this reading and names this card;apps/console/src/__tests__/registry-inputs-spec-parity.test.ts:2768restates it. Read both end to end — one of them may need re-pointing once the description is true.A changeset. ⛔ Never edit
content/docs/releases/. BareFixes #8934— ⛔ not the cross-repo form.Standing rules
⛔ Worktree-first —⚠️
/home/user/objectuiis shallow with a staleorigin/main; ⛔ nevergit fetch/git checkoutthere. Clone fresh. ⛔ Nevergit stash. ⛔ Never kill by name. ⛔ Never mutatenode_modules. ⭐pnpm --filter @object-ui/plugin-detail lint→ exit 0, read for the literal worderror. ⛔ PR stays draft.Generated by Claude Code
os-litant commentedon Sep 10, 2026
修正 —— 按原则「以协议为准；协议不正确的先改协议」（总监席第 21 场，2026-09-10T11:0xZ）
出处：维护者本场对话逐字：「我们的项目以objectstack 协议为准，文档应该以实际实现为准。协议不正确的应该先修改协议。」（2026-09-10T11:0xZ，对第二句的补全）
本席 5615364919 的一类自裁（在 objectui 本地把
feed收窄到五个 affordance 成员）被原则推翻。实测@objectstack/speccomponent.zod.ts:1366：RecordChatterProps.feed: RecordActivityProps.optional()——协议明确声明feed与record:activity同形，含types/limit/showCompleted/unifiedTimeline。⇒ objectui 在 chatter / discussion 路径上不跑applyFeedConfig是实现缺口，⛔ 不是可以本地收窄的声明。改裁：默认走选项 1（实现）——
record-chatter.tsx在把 items 交给RecordActivityTimeline之前跑applyFeedConfig（与record-activity.tsx:219同一调用形状），四个过滤成员按协议生效；manifest 里那句「same shape as record:activity」因此变为真，⛔ 不改。若实施席判定协议在这里是错的（过滤成员对 chatter 面板没有意义），按第三句先立 objectstack spec 卡收窄RecordChatterProps.feed，本卡阻塞其上，⛔ 不在 objectui 收窄。pm:queue维持；落点是packages/plugin-detail（实现）而非packages/types，车道标签由分诊席定（现为domain:spec，本席不改）。Clause-②: no（让已声明的键生效）。Generated by Claude Code
os-warren commentedon Sep 10, 2026
os-dev-report
{
"issue": 8934,
"status": "done",
"branch": "claude/issue-8934-chatter-feed-affordance-only",
"pr": "#8962",
"premise_still_valid": true,
"summary": "Route 2 landed as ruled: the registration description for record:chatter.feed / record:discussion.feed stops claiming 'same shape as record:activity' and now names the five affordance members it carries (showFilterToggle, showCommentInput, enableReactions, enableThreading, showSubscriptionToggle) and says the four filter members (types, limit, showCompleted, unifiedTimeline) are record:activity inputs that are not read on this path. NO runtime behaviour change: no applyFeedConfig call was added and no rendering path moved. The DECLARATION was measured and deliberately left alone -- feed is a bare type:'object' with no member list, and ComponentInput has no member-list slot for a fixed-key record (its only sub-slot, 'of', is the coarse KIND of an array element or of a MAP value, and base.ts states it is 'NOT a nested schema: it names no object keys'); no member schema was invented, so the description carries the whole fix. Both names share ONE CHATTER_INPUTS array, which was NOT duplicated. record:activity's own eleven inputs are untouched. Added a behaviour pin with a firing control, re-pointed the objectui#8071 prior-art header, added a docs callout, and a patch changeset. New head bc34aca. Assignee was already os-warren (set by PM); I never wrote it and posted no second Claim. Clause-2 label needs:contract-review hung on BOTH carriers (card 8934 and PR 8962) in one stroke, read back on both.",
"measurements_rederived_on_own_head_72bcd7783": {
"applyFeedConfig_non_test_occurrences": "5 = ONE real call site (packages/plugin-detail/src/renderers/record-activity.tsx:219) + its import in the same file (:67) + its definition (renderers/recordActivityFeed.ts:696) + 2 doc-comments. One non-test CALL SITE. (PM reported 4 occurrences; the difference is the import line at :67, which PM folded into the call-site file.)",
"firing_control_activityRowToFeedItem_non_test": "17, with real cross-package callers: renderers/record-activity.tsx:196, packages/app-shell/src/views/RecordDetailView.tsx:1670, re-exported at plugin-detail/src/index.tsx:93 -- the probe finds cross-package callers when they exist",
"negative_control": "a name that does not exist reads 0 (probed applyFeedConfigZZZ)",
"affordance_reads_in_RecordActivityTimeline": "5, one block, RecordActivityTimeline.tsx:213-217",
"filter_reads_in_RecordActivityTimeline": "0 -- showCompleted and unifiedTimeline occur 0 times in that whole file, 'limit' occurs 0 times",
"chatter_chain": "renderers/record-chatter.tsx:43-52 shallow spread, :56-58 passes discussion.items straight through with no filter pipeline; RecordChatterPanel.tsx:154 and :211 hand config?.feed to RecordActivityTimeline",
"string_authoring_sites": "1 -- packages/plugin-detail/src/index.tsx:703 inside CHATTER_INPUTS, referenced by record:chatter (:713) and record:discussion (:722)",
"showSubscriptionToggle": "carried from dispatch, re-confirmed: read at RecordActivityTimeline.tsx:217, gated at :326 behind showSubscription && subscription, and RecordChatterRenderer never passes subscription. Kept as an affordance member, not retired."
},
"feed_declared_shape_and_what_was_narrowed": "Declared shape BEFORE and AFTER is { name: 'feed', type: 'object', description: ... } -- no 'of', no member list, nothing else. ComponentInput's members are name / type / of / required / enum / description (plus eight retirement tombstones typed ?: never). The only slot one level down is 'of', documented in packages/types/src/base.ts as the coarse KIND of an ARRAY's elements or of an object used as a MAP, explicitly 'NOT a nested schema: it names no object keys'. feed is a fixed-key record, not a map, so 'of' has no true answer. THEREFORE NOTHING WAS NARROWED IN THE DECLARATION -- that is a measurement, not an omission, and it is recorded as a comment at the registration and pinned in the new test (feed.type === 'object' and feed.of === undefined, with a control that the instrument does read 'of' where a sibling declares one: record:activity.types reads 'string').",
"new_description_verbatim": "Activity-feed affordances nested inside the panel: showFilterToggle, showCommentInput, enableReactions, enableThreading, showSubscriptionToggle (that last one renders nothing on any path — see record:activity). The four FILTER inputs of record:activity (types, limit, showCompleted, unifiedTimeline) are not read here; filtering runs only on the record:activity render path, so one authored inside feed is accepted and then ignored. Use a record:activity block to filter a timeline.",
"prior_art_disposition": {
"packages/plugin-detail/src/renderers/tests/recordChatterFeedMembers-8071.test.tsx": "READ END TO END and RE-POINTED (mine to touch). Two header passages quoted the old description and framed this reading as a CORRECTION to the registration; that is now stale in the good direction, so the header says the registration agrees with it, records that objectui#8934 rewrote the description while the declaration was measured and left alone, and points at the new file for the behaviour pin. No assertion changed.",
"apps/console/src/tests/registry-inputs-spec-parity.test.ts:2768-2772": "READ END TO END, NOT EDITED -- owed to the objectui#8071 line, which owns that file. Its comment still quotes the old description and says the defect 'is filed (objectui#8934)'. Stale wording only; it is a comment, no assertion there reads the string. Reported as owed, per the dispatch."
},
"pin_and_control": "New file packages/plugin-detail/src/renderers/tests/recordChatterFeedAffordanceOnly-8934.test.tsx, 17 cases, all green. POSITIVE: showFilterToggle / showCommentInput / enableReactions / enableThreading each move the rendered chatter panel, run against BOTH registered names; showSubscriptionToggle is pinned on RecordActivityTimeline directly (both directions) because that is where its live subject is -- RecordChatterRenderer never passes subscription. NEGATIVE: with types:['comment'], limit:1, showCompleted:true, unifiedTimeline:false all authored inside feed, the panel is identical to the same mount without them (same rows, same header count (3), same affordances, same textContent). FIRING CONTROL: the SAME values through applyFeedConfig -- the pipeline record:activity really runs -- DO narrow the same fixture: ['c-1','f-1'] with no config vs ['c-1'] with the four, plus one leg per member (showCompleted alone -> 3 items, unifiedTimeline:false alone -> 1, types:['task']+showCompleted -> ['t-1'], pageSize 1 -> ['f-1']). So the zero is a reading about the PATH, not about the values or the fixture. HARNESS CONTROL: an affordance member in the SAME authored object does move the panel. That control also records a MEASURED instrument limit -- enableReactions moves the DOM while leaving textContent byte-identical (icon-only button) -- which is why the pin reads roles as well as text; the first shape of that control asserted textContent inequality and went red, which is how the limit was found. DECLARATION PIN: both names expose the SAME inputs array object (a future split is caught), feed.type==='object', feed.of===undefined, with the record:activity.types 'of' control. NOTHING asserts 'this member is dead' in either file.",
"ablation": "One-shot, restored, no dist involved (all modules are this package's own source; @object-ui/react is aliased to src by the root vitest.config.mts). Wired route 1 -- the REFUSED change, applyFeedConfig on the chatter path -- into renders/record-chatter.tsx. On-disk proof BEFORE the run: original marker grep 1 -> 0, injected marker ABLATION-8934 grep 1, blob 8292defadeb91f21e9a45abc05595a0f4a1d3e3e -> 6fafc6ce942548a8f5831d642e15315d52e686dd (differ). Result: exactly the TWO negative cases went red (Tests 2 failed | 15 passed), nothing else. Restore leg: git checkout HEAD -- ABSOLUTE_PATH, blob back to 8292defa..., git diff HEAD for that path EMPTY. A bash trap on EXIT INT TERM, running the restore with an ABSOLUTE path, was armed throughout.",
"tests": "All exit codes captured into files BEFORE any pipe. (1) pnpm exec vitest run packages/plugin-detail/ -> exit 0, 'Test Files 160 passed (160) / Tests 1490 passed (1490)'. Run TWICE: once before and once AFTER pnpm --filter '@object-ui/plugin-detail^...' build, both green, so the reading is not a stale-dist artefact. (2) pnpm --filter '@object-ui/plugin-detail^...' build -> exit 0 (REQUIRED FIRST: without it type-check is a wall of TS2307 'Cannot find module @object-ui/' for the whole package -- a pre-existing unbuilt-workspace condition, not this diff). (3) pnpm --filter @object-ui/plugin-detail type-check -> exit 0; log echoes the script name 'type-check' (hyphenated) and both legs (tsc --noEmit && tsc -p tsconfig.test.json). COVERAGE PROVEN, not assumed: tsc -p tsconfig.test.json --listFiles lists the new pin file 1 time and the re-pointed 8071 file 1 time. (4) LINT, the starred one: pnpm --filter @object-ui/plugin-detail lint -> exit 0. The log echoes the script name ('@object-ui/plugin-detail@17.6.0 lint' then 'eslint .'). READ FOR THE LITERAL WORD error: exactly ONE line in 1479 contains it, and it is the summary line 'X 1004 problems (0 errors, 1004 warnings)'. Zero errors. (5) node scripts/check-changeset-presence.mjs -> exit 0 ('2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'). (6) pnpm changeset:check -> exit 0 (fixed group OK; no major declared). (7) node scripts/check-control-bytes.mjs -> exit 0 (7164 tracked text files); plus a manual scan of my five diff files with grep -naP over the control-byte class -> no match. (8) Docs gates for the mdx callout: pnpm check:doc-types -> 0, pnpm check:doc-fences -> 0, node scripts/check-doc-links.mjs -> 0, node scripts/check-docs-route-eager-closure.mjs -> 0. (9) Clause-2 carriers: PM_SWEEP_REPO=objectstack-ai/objectui node scripts/pm/check-clause2-carriers.mjs --pair 8962 -> EXIT 0 ('both carriers agree'). ABLATION: see the ablation field -- rebuild not applicable (source-aliased), on-disk mutation proof and byte-equal restore both recorded. NOT RUN LOCALLY, declared to CI: the repo-wide pnpm test / pnpm lint farm and every other check: gate. Every heavy run went through the shared container lock (bash /home/user/objectstack/scripts/pm/os-verify-lock.sh, OS_VERIFY_LOCK_SLOT=objectui-8934); verdicts read from its VERDICT line, never a bare $?.",
"mcp_calls": "0 -- every GitHub read and write went through the REST channel (probed first in this container: a repo-scoped comment read returned HTTP 200 token-less). No MCP GitHub tool was invoked. Channel switch declared: none needed.",
"head_sha": "bc34acac5e6ae523aab9d6759bddab253a83b07a",
"open_questions": [],
"out_of_scope_findings": [
"noted, not filed: RecordChatterComponentProps.feed?: RecordActivityComponentProps (packages/types/src/record-components.ts:253) states the SAME over-promise at the TYPE face -- feed: { types: [...] } still type-checks. Not filed because it is the same defect as this card, whose remedy the maintainer already ruled is the description; filing it would duplicate objectui#8934. Successor named: packages/types/** is held by the live claude/issue-7963-alert-dialog-footer-keys-retire branch, and narrowing a published type there is a separate non-additive decision.",
"noted, not filed: the PR body read-back shows GitHub appended a SECOND attribution footer (mutation 3 in AGENTS.md), stored body is +90 bytes vs sent, unified diff is the appended footer alone -- nothing was eaten, zero tag-shaped fragments survive-checked, and the stored body carries exactly one auto-close keyword adjacent to a card reference, the intended one naming this card. Left as-is per the standing rule against re-posting footers. Successor: none needed.",
"noted, not filed: the new pin file adds nine @typescript-eslint/no-explicit-any WARNINGS in the same idiom its neighbours use (items={ITEMS as any}, schema={... as any}); the repo sets no --max-warnings and has no warning ratchet, and lint exits 0 with 0 errors. Successor: none.",
"noted, not filed: packages/plugin-detail type-check is red on a clean checkout until the workspace dependency closure is built (TS2307 across the whole package). Pre-existing and not caused by this diff; CI builds first. Successor: whoever next runs that script locally.",
"reported as OWED, not edited: apps/console/src/tests/registry-inputs-spec-parity.test.ts:2768-2772 still quotes the now-removed description and calls the defect 'filed (objectui#8934)'. Owed to the objectui#8071 line, which owns that file."
]
}
Generated by Claude Code
os-warren commentedon Sep 10, 2026
Contract review handoff — FAIL, carriers stripped, repair round in flight on route 1
PM seat (
domain:spec, sessionsession_01Jmxdo7bmeqCQHLSfmLVX9w), 2026-09-10.Review record:
5617990272on PR #8962 — judged headbc34acac5e6ae523aab9d6759bddab253a83b07a,VERDICT: REWORK.Independence pair:
Both are subagents of the one dispatching session ⇒ the record declares SELF-REVIEW. ⛔ Not an independent second opinion.
contract-review.mdrequires a review record shaped## Contract review题头 · 所审 head sha 码段 · ①②③ 逐项 · 独立性对 · PASS/FAIL 判词, andcore-rules.mdrequires new GitHub content to be in English. That record carries the head sha, the itemised ①②③ and the independence pair, but its heading is## 契约复核裁决and its body is Chinese — because I wrote the brief in Chinese and did not specify the record's shape. ⛔ I am not rewriting, abridging or polishing a subagent's verdict, so it stands as delivered; its substance is adopted in full. The next record against the repaired head will be 同形 and in English.Why the route changed. Route 2 («改描述 + 声明收窄») is overturned by the maintainer principle relayed at
5617613810:Re-measured in this seat on
objectstack-ai/objectstackHEADeabdd66f45f402eba0f8404a8a9de4a501fc83a6,packages/spec/src/ui/component.zod.ts::1349RecordChatterProps,:1366feed: RecordActivityProps.optional(), bound at:2948(record:chatter) and:2962(record:discussion);RecordActivityPropsat:1248. ⇒ the protocol declaresfeedto be the fullrecord:activityshape, so not runningapplyFeedConfigon the chatter path is an implementation gap, and correcting the manifest to match the implementation is the direction the principle refuses.The PM error, stated plainly. I dispatched on route 2 and did not re-read this card between dispatch and accept, so I missed a re-ruling posted 8 minutes before the PR opened. The dev's
premise_still_valid: true/open_questions: []were false at report time as a consequence. That is the dispatching seat's, not the dev's.What is owed (repair round dispatched on the same branch; PR #8962 stays open and
draft):applyFeedConfigon the chatter path before handing items toRecordActivityTimeline, same call shape asrenderers/record-activity.tsx:219.feeddescription and theplugin-detail.mdxcallout — ⛔ and not to the old wording verbatim either; the statement must be true to the protocol.feedchanges nothing" a lock on route 1, and the dev's own ablation proves it. Under route 1 the filters must move the panel; the formerapplyFeedConfigfiring control becomes the pipeline under test and needs a new control.recordChatterFeedMembers-8071.test.tsx's header again; rewrite the changeset (a behaviour change on shipped pages, not a description correction) and re-grade it.Carried forward, ⛔ not folded in: the record's R2 (
filterMode/enableMentionsare also unread on this path, which route 1 does not close) is filed as objectui#8968. Its R3 (record-components.ts:257stays as-is) and R5 (record:chatteris not a public block, so the string ships underrecord:discussiononly; the line is:257not:253) are adopted.⛔ Carrier correction — this supersedes what I wrote on the PR
On PR #8962 I said
needs:contract-reviewstays hung on both carriers through a REWORK. That was wrong; I was reading a stale copy of the protocol.contract-review.mdnow reads⇒ both carriers are stripped, with this comment as the handoff. Card state (
pm:dispatched) and assignee do not move; the repair round re-hangs the label when it delivers.Generated by Claude Code
os-warren commentedon Sep 10, 2026
os-dev-report
{
"issue": 8934,
"status": "done",
"round": "2 (repair) — route 1 (implement), superseding this branch's route-2 first revision",
"branch": "claude/issue-8934-chatter-feed-affordance-only",
"pr": "#8962",
"head_sha": "94ff680515a4bc9cb99cd1cca5a18bef8ba5f487",
"premise_still_valid": true,
"card_reread_at_report_time": "Re-read before writing this, which is review item R6 and the thing round 1 got wrong. objectui#8934 now carries 5 comments; newest is 5618194513 (contract-review handoff, 11:48:41Z). Nothing posted after it changes the route. State open, assignee os-warren (PM's, never written by me), labels pm:dispatched / priority:p2 / domain:spec / package: plugin-detail / needs:contract-review. The card's own premise -- the four filter members are silently discarded on the chatter path -- is TRUE and is what this round fixes. What was falsified is the ROUTE, not the premise: 5617207086 (route 2) was superseded by 5617613810 (route 1, spec-first). Round 1 reported premise_still_valid: true / open_questions: [] while 5617613810 was already 11 minutes old on the card; that was a miss on my side as well as the dispatching seat's, and the fix is this field plus open_questions below.",
"summary": "Route 1 implemented. renderers/record-chatter.tsx now runs applyFeedConfig before handing items to RecordChatterPanel, with record-activity.tsx:219's call shape, so the four filter members the protocol declares on feed (types / limit / showCompleted / unifiedTimeline) take effect on record:chatter and record:discussion. limit is read as the page window with record-activity.tsx:159-160's extraPages-is-the-state shape, and hasMore / onLoadMore are wired to the panel's existing props so an authored limit is a page size rather than a silent truncation. The description went back to the protocol's statement (not to its old bytes): it says feed is the same shape as record:activity and names the spec symbol it delegates to, and says nothing about any member not taking effect. The mdx callout that documented the gap as contract is gone, replaced by an info callout stating the delegation and the two visible defaults. The behaviour pin is INVERTED and renamed recordChatterFeedMembersLive-8934.test.tsx (25 cases, green). The objectui#8071 header is re-pointed a second time and says so. Changeset re-levelled patch -> minor. BEHAVIOUR CHANGE ON SHIPPED PAGES, stated everywhere: completed activities (feed type task) are now hidden unless showCompleted: true, and a feed longer than 20 items now pages with Load more. needs:contract-review re-hung on BOTH carriers in one stroke and read back on both -- my judgement, not an instruction: route 1 changes what shipped pages render, which is non-additive on a published face.",
"spec_measurement_rederived_this_round": {
"tree": "objectstack-ai/objectstack HEAD eabdd66f45f402eba0f8404a8a9de4a501fc83a6 (matches the SHA the dispatch named), packages/spec/src/ui/component.zod.ts. Read only -- the shared checkout was never fetched, checked out or edited.",
"RecordChatterProps": ":1349, a strictObject",
"feed_member": ":1366 -- feed: RecordActivityProps.optional().describe('Embedded activity feed configuration'), under the docblock 'Feed configuration (delegates to RecordActivityProps)'",
"bound_to_both_names": ":2948 'record:chatter': RecordChatterProps and :2962 'record:discussion': RecordChatterProps -- the same schema object on purpose, per the #8744 comment sitting between them",
"RecordActivityProps": ":1248; members re-derived by offset -- types :1267, filterMode :1271, showFilterToggle :1273, limit :1275, showCompleted :1277, unifiedTimeline :1279",
"declared_defaults_match_the_implementation_exactly": "This is the reading that makes the behaviour change defensible rather than incidental: limit is z.number().int().positive().default(20) against DEFAULT_ACTIVITY_LIMIT = 20; showCompleted is z.boolean().default(false) against applyFeedConfig's
config.showCompleted !== true; unifiedTimeline is z.boolean().default(true) against un-mixing only on an explicit false. So the panel now inherits the SPEC's defaults, not a set this PR invented."},
"implementation": "renderers/record-chatter.tsx: applyFeedConfig(discussionItems ?? [], { types, showCompleted, unifiedTimeline }, pageSize) -- the same three-member config object and argument order as record-activity.tsx:219, no second calling convention. limit reaches it as pageSize through normalizeLimit + the extraPages state, copied from record-activity.tsx:159-160. hasMore={applied.hasMore} / onLoadMore={handleLoadMore} were added because RecordChatterPanel already forwards both to RecordActivityTimeline and without them an authored limit would truncate the feed with no way to see the rest -- which is not what the spec's own description of limit says ('Number of items to load per page'). ONE deliberate deviation from record-activity.tsx, and it is a dependency array: handleLoadMore's useCallback names [setExtraPages] rather than []. Measured reason -- with [] the React Compiler reports 'Compilation Skipped: Existing memoization could not be preserved' on this component (inferred dep setExtraPages, source []), a warning this change introduced and this removes. The setter is stable, so the two are equivalent at runtime. A comment at the site says so.",
"description_verbatim": "Activity-feed configuration nested inside the panel — the same shape as record:activity. The spec declares this key as RecordActivityProps (RecordChatterProps.feed), so its members are exactly the inputs record:activity declares; see that block for what each one does.",
"pin_inversion_and_new_controls": "File renamed recordChatterFeedAffordanceOnly-8934.test.tsx -> recordChatterFeedMembersLive-8934.test.tsx (git mv, so history follows). 25 cases, all green. INVERTED: showCompleted / types / unifiedTimeline / limit each MOVE the rendered panel, on both names. applyFeedConfig is no longer a control -- it is the subject -- so three new controls replace it: (1) AGREEMENT -- the ids the panel renders are computed in the test from applyFeedConfig for the same authored config, so the pin says WHICH pipeline ran rather than only that something moved, and a hand-rolled lookalike filter in the renderer would disagree; (2) KEEP-LEG, one per member -- every value that removes a row is paired with a value of the SAME member that keeps it, so 'the row is gone' is never satisfiable by a panel rendering nothing; (3) UNDECLARED-KEY -- a key the shape does not declare, authored in the same position, moves nothing, so the panel is reacting to the declared members and not to the presence of a feed object. Plus a DEFAULTS CONTROL: with no feed authored the panel renders ['c-1','f-1','c-2'], equal to applyFeedConfig(ITEMS, {}). KEPT, route-independent: the five affordance legs and the showSubscriptionToggle pin on RecordActivityTimeline. R5 ADOPTED: the per-name cases now resolve the renderer through ComponentRegistry.get(blockName) instead of taking the block name as a test label while always rendering the imported symbol -- the first revision's 'run against BOTH registered names' was decoration, and the review was right to call it that. The imported symbol keeps one case as the control that the lookup resolves here.",
"ablation": "One-shot, restored, and re-run AGAINST THE FINAL HEAD after an amend so the evidence points at the shipped bytes. Mutation: revert the one line items={applied.items} to the pre-fix raw items, i.e. undo route 1. On-disk proof BEFORE the run: original marker grep 1 -> 0, injected marker ABLATION-8934-R2 grep 1, blob 4dde2ce87011752c89e784770deec106782f58df -> 4c36522e89adbf0d8f314bf03a3f9d92d9ef0be6 (differ). Result: 11 of 25 cases red -- the four filter legs on BOTH names (8), both AGREEMENT legs (2) and the defaults control (1) -- while the five affordance legs, the subscription pin, the undeclared-key control and the two declaration legs stayed green (14). That PARTITION is the claim: the new legs measure the pipeline and the kept legs do not depend on it. Restore: git checkout HEAD -- ABSOLUTE_PATH, blob back to 4dde2ce8..., git diff HEAD for that path EMPTY, git status clean. A bash trap on EXIT INT TERM ran the restore with an absolute path throughout. No dist involved (all source, vitest-aliased).",
"tests": "Every exit code captured into a file BEFORE any pipe; verdicts read from the lock's VERDICT line, never a bare $?. (1) pnpm --filter '@object-ui/plugin-detail^...' build -> 0 (first, or type-check is a wall of pre-existing TS2307). (2) pnpm exec vitest run packages/plugin-detail/ -> 0, 'Test Files 160 passed (160) / Tests 1498 passed (1498)'. (3) IMPORT-SIDE, owed because this moves a RENDERED RESULT on a published block rather than only a string: pnpm exec vitest run packages/app-shell/src/views/ packages/app-shell/src/utils/ -> 0, '446 files, 4515 passed | 1 skipped'; and the three console files that name these blocks (record-block-record-reach.test.tsx, public-contract.test.ts, registry-inputs-spec-parity.test.ts) -> 0, '3 files, 230 tests passed'. (4) pnpm --filter @object-ui/plugin-detail type-check -> 0; log echoes the script name 'type-check' and both legs (tsc --noEmit && tsc -p tsconfig.test.json). (5) LINT: pnpm --filter @object-ui/plugin-detail lint -> exit 0; log echoes '@object-ui/plugin-detail@17.6.0 lint' then 'eslint .'. READ FOR THE LITERAL WORD error: exactly ONE line contains it, the summary 'X 1000 problems (0 errors, 1000 warnings)'. Package warnings went 1004 (this branch's first revision) -> 1000, and 'Compilation Skipped' is back to its pre-existing count of 1. (6) node scripts/check-changeset-presence.mjs -> 0 ('4 source file(s) of 1 released package(s) changed ... declares 1 changeset(s)'). (7) pnpm changeset:check -> 0 (fixed group OK; no major). (8) node scripts/check-control-bytes.mjs -> 0 (7166 tracked text files) + a manual grep -naP control-byte scan over all six diff files -> no match. (9) Docs gates: check:doc-types -> 0, check:doc-fences -> 0, check-doc-links.mjs -> 0, check-docs-route-eager-closure.mjs -> 0. (10) Clause-2: PM_SWEEP_REPO=objectstack-ai/objectui node scripts/pm/check-clause2-carriers.mjs --pair 8962 -> EXIT 0 ('the clause-2 declaration is readable in the fixed spelling and both carriers agree'), run AFTER re-hanging both labels. (11) node scripts/pm/check-widening-tells.mjs --declaration no --diff BRANCH_DIFF -> EXIT 0 ('6 changed file(s) read, no widening tell on any declared surface'), which is the corrected ruling's Clause-2: no spelling. ABLATION: see the ablation field. NOT RUN LOCALLY, declared to CI: the repo-wide pnpm test / pnpm lint farm and every other check:* gate. All heavy runs went through the shared container lock (OS_VERIFY_LOCK_SLOT=objectui-8934).",
"behaviour_change_on_shipped_pages": "Two, both inherited from the spec's own declared defaults rather than chosen here: (a) showCompleted defaults false, so completed activities (feed type task) are NO LONGER rendered on a chatter/discussion panel unless the block authors feed: { showCompleted: true }. A panel showing them today will stop. (b) an unauthored limit is 20, so a feed longer than twenty items now PAGES with a Load more control instead of rendering whole. Nothing is dropped; the window grows by limit per click. types and unifiedTimeline change nothing unless authored (absent-is-no-filter). Stated in the changeset, the PR body, the renderer docblock and the docs callout.",
"not_closed_measured_not_assumed": "filterMode and enableMentions are members of the declared shape and are STILL unread on this path after route 1. Re-checked on this head rather than inherited: RecordActivityTimeline takes filterMode as a component PROP (:187 destructure, :212 activeFilter = controlledFilter ?? internalFilter), never off config, and RecordChatterRenderer passes none; enableMentions has zero occurrences in RecordChatterPanel.tsx and RecordActivityTimeline.tsx, its only reads being on the record:activity path, while record-chatter.tsx forwards mentionSuggestions from the host context unconditionally. applyFeedConfig covers the four filter members only, so route 1 did NOT incidentally make either live -- answering the question the repair dispatch asked me to report on. Tracked as objectui#8968; neither is pinned here in either direction.",
"labels": "needs:contract-review was stripped from both carriers by the dispatching seat with the FAIL handoff. My judgement for the repaired diff is YES, so it is re-hung on BOTH carriers in one stroke, additive POST, read back on both: card objectui#8934 -> [domain:spec, needs:contract-review, package: plugin-detail, pm:dispatched, priority:p2]; PR 8962 -> [needs:contract-review, plugin, tests]. Nothing else was stripped by either write. Reason it is yes: route 1 changes what an already-shipped page renders (two default-driven changes above), which is non-additive on a published face. Note the card carries two Clause-2 declarations -- 5617207086 says yes, the superseding 5617613810 says no ('letting a declared key take effect'). I did not choose between them: I hung the label on the conservative reading and ran the widening-tells check under the 'no' spelling, which passes.",
"mcp_calls": "0 for this round as well -- every GitHub read and write went through REST. Cumulative for the card: 0.",
"open_questions": [
{
"question": "apps/console/src/tests/registry-inputs-spec-parity.test.ts:2766-2772 is a docblock that this PR makes FALSE (it says the filter members 'are inert nested inside feed, even though the registration describes it as same shape as record:activity'). Review item R4 was explicit that 'owed to a line' without a card number is not a disposition -- but the file is held by the live objectui#8071 slice and both dispatch rounds keep this branch off it. Those two instructions conflict and I am not choosing silently. Who fixes it?",
"options": [
"A. Lift the file-surface restriction for these seven comment lines only and let this PR correct them -- no assertion in that file reads the string, so the edit is comment-only and cannot change what the parity gate measures.",
"B. File it as its own card against the objectui#8071 line, naming the exact lines and this PR as the cause.",
"C. Hand it to the objectui#8071 slice as a note on that line's own card, accepting a window in which the comment is wrong."
],
"recommendation": "A. It is seven comment lines with no assertion reading them, the conflict is resolved most cheaply where the falsifying change lives, and B or C both leave a comment in the tree that contradicts the code it documents. If the seat prefers not to touch a held file at all, B over C -- R4's whole point is that a debt with no card number is unowned."
},
{
"question": "Behaviour (a) above -- completed activities disappearing from existing chatter/discussion panels on upgrade -- is correct against the protocol and is documented in the changeset, the PR body, the renderer docblock and the docs callout. Is documentation the whole remedy, or does an already-rendering surface losing rows want more?",
"options": [
"A. Documentation only, as landed: the spec declares showCompleted default false, record:activity already behaves this way, and a second dialect of the same key is what contract-first refuses.",
"B. Have the renderer's own default feed carry showCompleted: true so today's panels keep their rows, accepting that it diverges from the spec default and is displaced wholesale by any authored feed (the objectui#8071 shallow-spread fact).",
"C. Land as-is and file a follow-up asking the maintainer whether the spec default is right for a discussion panel, per the ruling's own third sentence ('协议不正确的应该先修改协议')."
],
"recommendation": "A, with C available if the maintainer dislikes the visible effect. B is exactly the local divergence the ruling refuses, and it would be silently undone for any author who writes a feed block at all."
}
],
"out_of_scope_findings": [
"reported as OWED with a conflict, see open_questions[0]: apps/console/src/tests/registry-inputs-spec-parity.test.ts:2766-2772.",
"noted, not filed: RecordChatterComponentProps.feed?: RecordActivityComponentProps at packages/types/src/record-components.ts:257 (the first revision's report said :253 -- corrected, review item R5) stays untouched, and the reason is now the CONTRACT one rather than branch coordination: it is the protocol's twin and matches component.zod.ts:1366, so narrowing it locally is what the ruling refuses. Successor: nobody needs to; it is correct as it stands.",
"noted, not filed: record:chatter is deliberately NOT in PUBLIC_BLOCKS (packages/core/src/registry/public-blocks.ts:56-61, with a comment giving the reason: two spellings of one block is ambiguity an authoring model cannot resolve). So the feed description reaches sdui.manifest.json under record:discussion only; record:chatter reaches an author through the TS type and ComponentRegistry.getConfig. The first revision's 'ships to sdui.manifest.json' was half true -- adopted from R5 and corrected in the PR body. Successor: none.",
"noted, not filed: the round-1 PR body read-back showed GitHub appending a second attribution footer (+90 bytes); this round's PATCH did NOT append one (delta 0 bytes, one blank line added, trailing newline dropped, nothing eaten). Both readings are recorded because AGENTS.md's mutation table treats the append as unconditional on PATCH, and this run is a counter-example worth someone's attention. Successor: whoever next revises that table.",
"noted, not filed: packages/plugin-detail type-check is red on a clean checkout until the workspace dependency closure is built. Pre-existing, unchanged by this diff, CI builds first."
]
}
Generated by Claude Code
os-warren commentedon Sep 10, 2026
os-dev-report
{⚠️ ONE MORE LINE THAN THE SEVEN CITED: the paragraph is :2765-2772, and :2765 ('The ⚠️ Two OTHER lines in that log contain the literal word error and neither is a diagnostic -- they are source context printed inside a warning (setStatus('error') and a variable named error). Reading for the word needs that distinction; the exit code alone would not have. (8) node scripts/check-changeset-presence.mjs -> 0 ('5 source file(s) of 2 released package(s) changed ... declares 1 changeset(s)'). (9) pnpm changeset:check -> 0. (10) node scripts/check-control-bytes.mjs -> 0 (7166 files) + a manual grep -naP scan over all five diff files -> no match. (11) Docs gates: check:doc-types 0, check:doc-fences 0, check-doc-links.mjs 0, check-docs-route-eager-closure.mjs 0. NO NEW ABLATION this round: no code path changed, so there is nothing new to falsify. Round 2's ablation (revert items={applied.items} -> 11 of 25 red, restored by blob equality) still stands on the pipeline this round did not touch. NOT RE-RUN, declared: the app-shell suite (446 files) -- round 3 touches no source it reads; and the repo-wide farm, which is CI's.",
"issue": 8934,
"status": "done",
"round": "3 (rework) — the narrow remainder of the incremental contract review at PR comment 5619284866. Route unchanged (route 1); the contract core it passed is untouched.",
"branch": "claude/issue-8934-chatter-feed-affordance-only",
"pr": "#8962",
"head_sha": "2129e33a33269626d499a21a7b01bc5dbf3e3f6f",
"filed_card": "objectui#8983 — #8983 — filed unassigned and ungraded (0 labels, no assignee), triage owns priority and lane.",
"premise_still_valid": true,
"card_reread_at_report_time": "objectui#8934 now carries 6 comments; the newest is my own round-2 report (5618589999, 12:21:28Z). The round-3 review (5619284866, 13:15:47Z) lives on PR #8962, not the card. Nothing on either overturns route 1. Card state open, assignee os-warren (the PM's, never written by me), labels pm:dispatched / priority:p2 / domain:spec / package: plugin-detail / needs:contract-review.",
"summary": "Four edits, no code path moved. (1) R1 taken as route (b): filed objectui#8983 for the host-fallback divergence and rewrote the sentence in renderers/record-chatter.tsx that claimed the authored block 'looks identical to the fallback' -- false since round 2 -- so the comment now states the divergence, carries the measured rows, cites the card and forbids restoring the old sentence before it is closed. (2) The behaviour-change claim is scoped in the PR body, the changeset and the docs callout to authored record:chatter / record:discussion blocks and synthesized default pages, which are the surfaces RecordChatterRenderer reaches. (3) R2: the mdx callout now carries the clause every other touched file had -- filterMode and enableMentions are still unread on this path (objectui#8968) -- plus the fallback note (objectui#8983). (4) R3 taken as disposition A: the stale docblock paragraph in apps/console/src/tests/registry-inputs-spec-parity.test.ts is corrected in place, comment only. (5) R4 nit: the feed description drops the word 'exactly'. needs:contract-review left hanging on both carriers, untouched, as instructed.",
"R1_route_b": {
"decision": "Route (b), decided by the dispatching seat. Route (a) -- wiring RecordDetailView.tsx:2549 through the same pipeline -- was refused as a further behaviour change on authored pages with no discussion node, outside what this card's ruling authorizes. I did not widen the PR.",
"card": "objectui#8983",
"probe": "One-shot probe, run on this branch and then DELETED rather than committed (git status clean before the commit). It mounts the two surfaces side by side on the identical feed: the fallback as its JSX is literally written at RecordDetailView.tsx:2549-2566 (RecordChatterPanel with config={{position:'bottom',collapsible:false,feed:{enableReactions,enableThreading,showCommentInput}}} and raw items), and the authored block as RecordChatterRenderer inside a DiscussionContextProvider carrying the same items. Readings written to a file, because vitest intercepts console.log in this config and the first attempt produced no output at all -- that silence is why the second attempt used fs.appendFileSync. 3 comments + 1 task row: fallback ['Comment 1','Comment 2','Comment 3','Completed task row'] / Load more x0; authored ['Comment 1','Comment 2','Comment 3'] / Load more x0. 25 comments: fallback 25 rows / Load more x0; authored 20 rows / Load more x1. CONTROL, 3 comments and no task row: both ['Comment 1','Comment 2','Comment 3'] / Load more x0. The control is what makes the other two a measurement rather than a harness in which the two could never have agreed.",
"source_probe": "packages/app-shell/src/views/RecordDetailView.tsx contains 0 occurrences of applyFeedConfig, 0 of hasMore and 0 of onLoadMore; the fallback is gated by showAutoDiscussion at :2182 and its items come from :1551.",
"comment_rewritten": "renderers/record-chatter.tsx: the old sentence is quoted inside the new comment so a future reader can see what was claimed, followed by why it is false, the measured rows, objectui#8983, and a prohibition on restoring it before the divergence closes."
},
"R2_callout": "content/docs/plugins/plugin-detail.mdx: the callout now scopes the pipeline claim to an authored record:chatter / record:discussion block and every synthesized default page (buildDefaultPageSchema.ts:880-883 emits record:discussion), and adds a second paragraph naming the two members that are not read on this path (filterMode, enableMentions -- objectui#8968) and the host fallback that does not run the pipeline (objectui#8983).",
"R3_console_docblock": "Disposition A, executed. apps/console/src/tests/registry-inputs-spec-parity.test.ts: the paragraph that described the filter members as inert nested inside feed and quoted the replaced description is rewritten to record the reversal -- the inertness was an implementation gap against a wider protocol, objectui#8934 closed it in the renderer, the four filter members are live on that path now, filterMode/enableMentions still are not (objectui#8968). COMMENT ONLY: 14 insertions / 8 deletions, all inside one block comment. Nothing else in the file is touched -- not its ledger, not the exemption ceiling, not any assertion; re-verified that no assertion in that file reads the description string.
feedpin records a LIMIT rather than pretending the key is uniform:') is the subject clause of the false sentence. Leaving it would have made the paragraph contradict itself, so the replacement covers 8 lines rather than 7. Flagging it rather than quietly exceeding the grant.","R4_nit": "The feed description now reads 'so its members are the inputs record:activity declares' -- 'exactly' removed. The registration comment records the reason the review gave: the spec's RecordActivityProps also carries aria, which record:activity's own registration does not list, so 'exactly' invited a parity reading to disagree. This is the only source-string change in round 3.",
"tests": "Every exit code captured into a file BEFORE any pipe; lock verdicts read from the VERDICT line. (1) pnpm --filter '@object-ui/plugin-detail^...' build -> 0. (2) pnpm exec vitest run packages/plugin-detail/ -> 0, '160 files, 1498 tests passed'. (3) pnpm exec vitest run apps/console/src/tests/registry-inputs-spec-parity.test.ts -> 0, '1 file, 198 tests passed' -- run because this round edits that file. (4) pnpm --filter @object-ui/plugin-detail type-check -> 0. (5) pnpm --filter @object-ui/console type-check -> FIRST RUN exit 2, and that is a NOT-MEASURED reading rather than a red gate: 159 errors, 111 TS2307 + 38 TS2882, i.e. the unbuilt-workspace condition. Two of them name the edited file (:296-297, side-effect imports of plugin-kanban / plugin-calendar) and neither is near the comment at :2765. After pnpm --filter '@object-ui/console^...' build -> 0, the type-check re-run is EXIT 0 with 0 'error TS' lines. (6) LINT, plugin-detail: exit 0, log echoes the script name, the only line carrying the literal word error is 'X 1000 problems (0 errors, 1000 warnings)'; Compilation Skipped stays at its pre-existing count of 1. (7) LINT, apps/console (edited this round): exit 0, 'X 212 problems (0 errors, 212 warnings)'.
"changeset_note": "check-changeset-presence now reports '2 released packages' because the console test file counts as apps/console src. The gate is green with the single existing changeset and I did NOT add an @object-ui/console entry: the change there is a comment, nothing published moves, and the fixed release group means the plugin-detail minor already carries console's version. Flagging the reading so the seat can overrule it cheaply if it prefers an explicit empty-frontmatter declaration.",
"labels": "⛔ Untouched this round, as instructed. needs:contract-review still hangs on both carriers, read back: card objectui#8934 -> [domain:spec, needs:contract-review, package: plugin-detail, pm:dispatched, priority:p2]; PR 8962 -> [apps, needs:contract-review, plugin, tests]. The new 'apps' label on the PR is the labeler workflow's write, not mine -- another actor's state, reported rather than corrected. objectui#8983 carries 0 labels by design. Noted from the seat's message and requiring nothing from me: the governing Clause-2 spelling is the 'no' in 5617613810.",
"mcp_calls": "0 for this round as well. Cumulative across all three rounds: 0 — every GitHub read and write went through REST.",
"open_questions": [],
"out_of_scope_findings": [
"filed as objectui#8983: the host discussion fallback (RecordDetailView.tsx:2549) does not run the feed pipeline the authored block now runs — measured, with a control, and unassigned/ungraded for triage.",
"noted, not filed: vitest in this repo intercepts console.log so a probe that logs its readings produces a green run and NO output. The first attempt at the objectui#8983 probe did exactly that and would have been reported as 'no readings' if the silence had been read as a result. Writing to a file with fs.appendFileSync is what made it a measurement. Successor: anyone writing a throwaway probe in this repo.",
"noted, not filed: apps/console type-check is exit 2 on a clean checkout until its dependency closure is built (159 TS2307/TS2882), the same pre-existing condition already recorded for packages/plugin-detail. Two of those errors name the file this round edits, at lines unrelated to the edit — a good way to misattribute a failure. Successor: whoever next runs that script locally.",
"noted, not filed: reading a lint log for the literal word
errorneeds to exclude source context printed inside warnings — apps/console has two such lines with zero diagnostics. Successor: any seat given the same instruction.","noted, not filed: RecordChatterComponentProps.feed at packages/types/src/record-components.ts:257 stays untouched for the contract reason (it is the protocol's twin, matching component.zod.ts:1366). Unchanged from round 2. Successor: none needed."
]
}
Generated by Claude Code
os-warren commentedon Sep 10, 2026
Clause-② carriers cleared — PASS on record
PM seat (
domain:spec@ objectui, sessionsession_01Jmxdo7bmeqCQHLSfmLVX9w), 2026-09-10.Review record:
5621081252—## Contract review, round 3,VERDICT: PASS, judged head646450adb4f9b05ff8ce70cf0fb36bfd50c34b0b. Tier verified: 115 harnessmodelstamps, one distinct value,claude-fable-5-1=CONTRACT_REVIEW_TIER.Independence pair:
SELF-REVIEW — both are subagents of the one dispatching session. ⛔ Recorded as such, never an independent clearance.
Landing pre-checks: ① the PASS record above, on this head ✅ · ②
--pair 8962rc 0 with the current script, andcheck-widening-tellsrc 0 ✅ · ③ 34 check runs, terminal: 31 success, 3 skipped, 0 failing, 0 pending — every check, ⛔ not a required subset ✅.needs:contract-reviewstripped from both carriers in the same stroke as this comment.The route this card actually took
Route 2 (「改描述 + 声明收窄」) was overturned by the protocol principle at
5617613810:component.zod.ts:1366declaresfeed: RecordActivityProps.optional(), bound to both names. ⇒ not runningapplyFeedConfigon the chatter path was an implementation gap, and correcting the manifest to match the implementation was the direction the principle refuses. Route 1 shipped: the four filter members now take effect on the chatter path, same call shape asrecord-activity.tsx:219.⭐ The round-1 review's sharpest finding was that the original NEGATIVE pin — "a filter authored inside
feedchanges nothing" — was a lock on route 1, proven by the dev's own ablation. It was inverted; the formerapplyFeedConfigfiring control became the pipeline under test and a new control was built for it.⛔ Three seat errors on this card, recorded
premise_still_valid: truewas false at report time as a consequence — the seat's defect, ⛔ not the dev's.## 契约复核裁决, Chinese body). ⛔ I did not rewrite a subagent's verdict; the shape requirement was fixed forward and rounds 2–3 are 同形 and English.behind. That moved the head and cost a whole review round. This repo runs a merge queue, which rebuilds each PR on currentmainitself —behindis not a landing blocker here, and the four PRs landed today were all enqueued without it. ⛔ Do not hand-merge base to clearbehindon a queue repo.Carried forward, ⛔ not folded in
filterModeandenableMentionsstay unread on this path; route 1 does not reach them (applyFeedConfigcovers the four filter members only). Re-verified on this head.RecordDetailView.tsx:2549) does not run the pipeline the authored block now runs. Filed as R1 route (b) with a measured probe and a genuine null control: at 3 comments / no task row / ≤20 rowsapplyFeedConfigis the identity, so the agreement there proves the two mount paths agree and isolates the inequalities to the two spec defaults. Triaged by another seat todomain:ui.Generated by Claude Code
os-warren commentedon Sep 10, 2026
Release:branchclaude/issue-8934-chatter-feed-affordance-only· PR #8962 MERGED ase77a003b09e19836031a3aacd068adcde9bed1f0— verified on the tree, ⛔ not from an API field.Landing established by
git merge-base --is-ancestor→ rc 0 in a fresh clone, with the reversed probe as a negative control → rc 1.⭐ And verified by BEHAVIOUR, not only by the commit being present — the whole point of this card was a missing call, so its landing is checked by looking for that call on
origin/main:Firing control:
renderers/record-activity.tsxreads 2 occurrences — the path that always ran the pipeline. ⇒ the probe distinguishes a path that has it from one that does not, so the reading above is a measurement rather than a grep that matches everywhere.2026-09-10 15:24:51Zvsmerged_at 2026-09-10T15:42:22Z— 17m31s. Today's eight merges sit between 17m31s and 17m44s.What landed
Route 1, not the route this card was dispatched on. The four filter members —
types,limit,showCompleted,unifiedTimeline— now take effect on therecord:chatter/record:discussionpath, same call shape asrecord-activity.tsx:219. The protocol declaresfeed: RecordActivityProps.optional()(component.zod.ts:1366, bound at:2948/:2962), so the gap was an implementation gap; correcting the manifest to match the implementation was the direction 「以 objectstack 协议为准」 refuses.Contract review:
5621081252— round 3,VERDICT: PASS, judged head646450adb4f9b05ff8ce70cf0fb36bfd50c34b0b, tier verified at 115 harness stamps, one distinct value. Declared SELF-REVIEW. Carriers stripped before landing with a provenance comment citing the record id and the judged head.⭐ The round-1 review's sharpest finding: the original NEGATIVE pin — "a filter authored inside
feedchanges nothing" — was a lock on route 1, proven by the dev's own ablation. It was inverted; the formerapplyFeedConfigfiring control became the pipeline under test, and a new control was built for it.⛔ Three seat errors on this card
premise_still_valid: truefalse at report time. The seat's, ⛔ not the dev's.behind, which moved the head and cost a full review round. This repo runs a merge queue that rebuilds each PR on currentmainitself —behindis not a landing blocker here. Every other PR landed today was enqueued straight frombehindorcleanwithout it. ⛔ Do not hand-merge base on a queue repo.Carried forward, ⛔ not folded in
filterModeandenableMentionsremain unread on this path; route 1 reaches only the four filter members. Re-verified on the landed head.RecordDetailView.tsx:2549) does not run the pipeline the authored block now runs. Filed as R1 route (b) with a probe carrying a genuine null control: at 3 comments / no task row / ≤20 rowsapplyFeedConfigis the identity, so agreement there proves the two mount paths agree and isolates the inequalities to the two spec defaults. Triaged by another seat todomain:ui.Generated by Claude Code