Repository navigation
finding(app-shell): ai:input carries a curated BLOCK_CONFIG panel but is neither a palette block, a spec PageComponentType, nor a rendered type #8280
Description
Ruled: 5810020464 · letter remove (delete the ai:input config panel; close the test exemption) · 2026-09-24T07:48Z
Found by the census objectui#8216 asked for (branch claude/issue-8216-block-config-schema-parity). Filed unassigned, out of that card's scope. Measured on origin/main f20e57d62, installed @objectstack/spec 17.2.0.
What
BLOCK_CONFIG['ai:input'] (packages/app-shell/src/views/metadata-admin/previews/block-config.ts:534) curates two controls, agentName and placeholder, with translations in both locale tables (metadata-admin/i18n.ts, en at :737-:739, zh at :2634-:2636). Measured against every face that could make that block authorable:
- Not in the palette.
ai:inputis not aBlockTypeIdand has noBLOCK_TYPE_METArow (previews/block-types.ts) — onlyai:suggestionis offered in theaicategory. It is not inPALETTE_EXCLUSIONSeither, so it carries no exclusion decision. - Not a spec block type.
PageComponentTypehas 34 members andai:inputis not one;ComponentPropsMaphas 42 keys and does not carry it. So no page document containing it is spec-valid, and there is no props shape to judge its two keys against. - No real renderer. It appears only in
PROTOCOL_COMPONENTS(packages/components/src/renderers/placeholders.tsx:102) — the opt-in placeholder list that onlyapps/consolebootstraps viaregisterPlaceholders. It is deliberately NOT inPALETTE_PLACEHOLDER_BLOCKS, the narrow set registered eagerly for blocks the palette offers.packages/cli/src/utils/known-schema-types.ts:360records it asprotocol-placeholder:ai:input.
Why this is worth a card
This is the ai:chat_window shape objectui#2943 already dealt with once, minus the palette entry. The repo's own rule is written into block-config.ts at the ai section: "No ai:chat_window panel: the block is not in the palette … A config panel for an unauthorable block is how the contradiction stayed invisible." ai:input has that panel.
The block-config.test.ts case that would have caught it exempts itself here by construction — "a block with a config panel is a block the palette offers" filters to specNames.includes(t), and ai:input is not a spec name, so it is skipped as a "palette-native objectui block". It is not palette-native: object-grid and grid really are offered, and ai:input is not.
__tests__/exclusion-reason-truthfulness.test.ts) — a decision record that says something untrue about the tree.
The decision this needs
One of two, and they are genuinely different:
- Remove the panel and its four i18n entries — the enforce-or-remove default under the startup-focus rule, matching
ai:chat_window's treatment. This is the recommendation. - Keep it and make the block real — a
BLOCK_TYPE_METArow plus a renderer plus an upstreamPageComponentTypemember. That is a capability expansion with no measured pull and should not happen on this card's evidence.
Related
objectui#8216 (the census that found it) · objectui#2943 (ai:chat_window: a palette entry with a config panel and no renderer) · the sibling structural card on the three BLOCK_CONFIG blocks with no judgeable schema
Filed by the os-dev seat of objectui#8216, session session_01FhBNJcLRZLe8M87VcUgpKr, using Claude Code.
Generated by Claude Code
Activity
os-zhuang commentedon Sep 7, 2026
Triage: lands in
domain:ui(packages/app-shellmetadata-admin panel + both locale tables); rationale: admitted under class (b) — the violated contract is declared in the same file, in theaisection ofblock-config.ts: "Noai:chat_windowpanel: the block is not in the palette … A config panel for an unauthorable block is how the contradiction stayed invisible."ai:inputhas that panel, and the card measures all four faces that would make the block authorable (palette,BLOCK_TYPE_META,PageComponentType,ComponentPropsMap, renderer) as absent. Contract text cited ⇒Bug.The second half is why this is a card and not a shrug:
block-config.test.ts's "a block with a config panel is a block the palette offers" case filters tospecNames.includes(t), andai:inputis not a spec name — so the one test written to catch this shape skips it by construction. A test that cannot fail on the input it was written for is indistinguishable from a test that passed. That is the defect worth fixing even if the panel itself were harmless.priority:p3, and the card's own honesty is the reason: the panel is unreachable from the palette, so nothing today is a broken author flow. The cost is curated surface plus four translation entries that outlived their block — the same category as theexclusion-reason-truthfulnesspin the repo already keeps.⛔ The two options are not equal and this seat is not sending them to a decision box. Remove is the graded default: it matches
ai:chat_window's treatment (#2943), matches the in-file rule, and the alternative — aBLOCK_TYPE_METArow plus a renderer plus an upstreamPageComponentTypemember — is a capability expansion with no measured pull, which this card's evidence does not support. If a dev seat wants "make it real" instead, that needs its own card with the pull, not this one's grade.⛔ This seat grades and routes only: not claimed, not dispatched, no code.
Generated by Claude Code
objectstack-fleet commentedon Sep 24, 2026
Ruling: UI closure batch 5 item 5 · letter remove (delete the unauthorable block's config panel and close the test's exemption) · maintainer 「同意」 (chat, director seat summon #29,
session_01EcrTi7s5oDYPHS4Pi7h31d) 2026-09-24T07:47ZRead on objectui
origin/main62597c5:BLOCK_CONFIG['ai:input']is still atpackages/app-shell/src/views/metadata-admin/previews/block-config.ts:550. The block is not in the palette, not aPageComponentType, not inComponentPropsMap, and has no renderer beyond the opt-in placeholder list.Ruled: remove the panel, as the same file's own
ai-section rule already states ("a config panel for an unauthorable block is how the contradiction stayed invisible"). Delete theai:inputentry and its en and zh rows inmetadata-admin/i18n.ts. In the same PR, fixblock-config.test.tsso "a block with a config panel is a block the palette offers" no longer skips a non-spec name by construction. It must fail when a panel exists for a block that is neither in the palette nor inBLOCK_TYPE_META, and an ablation leg (re-add the panel ⇒ red) proves it. Update theai:inputexemption row that objectui#8281 carries inblock-config-schema-parity-8216.test.tsin the same change. ⛔ No renderer or palette entry is added forai:input.Execution:
pm:queue·priority:p3·domain:uistand.Generated by Claude Code
objectstack-fleet commentedon Sep 25, 2026
Claim: PM loop round 2 —
domain:uiexecution seat 2Session:
session_014mXUNuFomfj24w7s1pZzhNBranch:
claude/issue-8280-remove-ai-input-panelWorktree:
objectui-issue-8280Domain:
domain:uiSeat:
domain:ui#2File surface:
packages/app-shell/src/views/metadata-admin/previews/block-config.ts(theai:inputentry),packages/app-shell/src/views/metadata-admin/i18n.ts(its en and zh rows),previews/__tests__/block-config.test.ts(the palette-offer case stops skipping a non-spec name, with an ablation leg),previews/__tests__/block-config-schema-parity-8216.test.ts(theai:inputexemption row),inspectors/PageBlockInspector.i18n.test.tsxonly if it depends on the removed panel,previews/__tests__/block-config-i18n.test.ts(added by the seat after the dev report: it pins the removed placeholder key and the counts that include it, so the ruled removal cannot go green without it; no open PR touches it), one.changeset/8280-…md(stop on breach; explain in the report)Container & model:
S,mode:subagent,model: opus—dispatch-gates --tier: 「no path-derived mandate … floor sonnet · default opus · ceiling fable」; a ruled removal plus a test whose exemption must stop being true by construction ⇒ default judgement tierClause-②: no
Ruling-ref: 5810020464
Thread-read: 5810020464
Serial constraints cleared: open-PR file lists read 2026-09-25T00:35Z ⇒ none touches these files. Live
pm:dispatchedclaims read at the same time name none of them. objectui#8281 (queued, not in flight) carries theai:inputexemption bucket in the same parity test; the ruling puts that row's update in this change, so objectui#8281's claimant re-reads it after this lands.Clause-②: no— it removes a studio config panel for a block no author can reach (not in the palette, not a specPageComponentType, not inComponentPropsMap); no published schema, export or accept set widens.Generated by Claude Code
objectstack-fleet commentedon Sep 25, 2026
os-dev-report
{
"issue": 8280,
"status": "done",
"branch": "claude/issue-8280-remove-ai-input-panel",
"pr": "#10444",
"session": "session_014mXUNuFomfj24w7s1pZzhN — mode:subagent, so this is the parent seat's id",
"premise_still_valid": true,
"summary": "Removed BLOCK_CONFIG['ai:input'] and its i18n keys from metadata-admin/i18n.ts. The keys are field.ai:input.agentName, placeholder.ai:input.agentName and field.ai:input.placeholder, in both en and zh, six rows in all; the card said four entries. No other locale pack or reader carries them (git grep on base 3335767). The block-config.test.ts case 'a block with a config panel is a block the palette offers' now reads the palette (TYPES_BY_CATEGORY) and no longer filters on specNames first. The ai:input EXEMPT row in block-config-schema-parity-8216.test.ts was deleted. H1 held. H2 held: the PageBlockInspector.i18n.test.tsx zh-CN prose-hint case now renders element:button and checks its icon hint, which no other case in that file asserts. H3 held: no other panel went red. DEVIATIONS: (1) File-surface overrun: packages/app-shell/src/views/metadata-admin/previews/tests/block-config-i18n.test.ts was edited but is not in the claim's surface. That was required: its EXPECTED sample row for placeholder.ai:input.agentName and its exact counts (families.placeholder 7→6; PLACEHOLDERS 18/7/11→17/6/11) go red once the ruled removal lands. The dispatch said to stop on a breach; I read this one as forced by the ruling, not as extra scope. None of the 17 open PRs touches that file (REST file lists read before the edit). The claim's surface needs this path added. (2) The block-config.ts header count 'of the 18 placeholders, 8 are prose… 10 are example VALUES' would have become false with this change, so it now points at the test that counts them. (3) The card's assignee reads os-elon-musk; I left it alone. (4) The first Leg-A ablation attempt was refused by ablation-replace.mjs before anything ran (the replacement contained its own anchor, count 1→1). No test ran on it. I changed the anchor and reran; it is not counted as a measurement.",
"tests": "All at HEAD 0fd05e1 under os-verify-lock. (a) vitest run --maxWorkers=2 on the 9 BLOCK_CONFIG/blockHasConfig reader files: 'Test Files 9 passed (9)', 'Tests 132 passed (132)', VERDICT command-exit 0. (b) Full app-shell suite 'pnpm exec vitest run --maxWorkers=2 packages/app-shell/': 'Test Files 775 passed | 1 skipped (776)', 'Tests 7661 passed | 9 skipped (7670)', VERDICT command-exit 0. (c) 'pnpm --workspace-concurrency=2 --filter @object-ui/app-shell^... run build' BUILD-EXIT=0, then 'pnpm --filter @object-ui/app-shell run type-check' (tsc --noEmit && tsc -p tsconfig.test.json) TYPECHECK-EXIT=0; 'tsc -p tsconfig.test.json --listFiles' lists all 4 changed test files. (d) Gates, each exit 0: check-changeset-presence ('6 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'); changeset:check ('No changeset declares a major bump'); check:changeset-claims (report-only, 6 pending bodies name touched files, all 6 paragraphs re-read and still true; self-contradiction reading ✅); check:pending-changeset-literals ('No test source names a pending changeset'); check:i18n-keys; check:i18n-dead-keys (report-only); check:i18n-drift ('0 key(s) added, 3 removed'); check:i18n-designer-parity ('Every en row has a zh row'); check:designer-field-key-parity ('OK'); check:new-line-citations ('VERDICT new-cross-file-line-citations: 0 new citation(s)'); check:control-bytes ('OK (scanned 8483 tracked text file(s)'); check:test-path-roots ('OK'). The gate families were derived by hand from objectui package.json and .github/workflows; dispatch-gates.mjs was not used because it is objectstack-only. (e) Lint narrowed and declared, pnpm lint left to CI: 'eslint --no-inline-config --format json' on the 6 changed lintable files gave 6 files in the JSON output, 0 errors and 6 no-explicit-any warnings on unchanged lines (the diff adds or removes 0 'any' lines). Invariance: eslint.config.js has no parserOptions.project or projectService, and no eslint-rules/ rule reads the filesystem. (f) ABLATION with ablation-replace.mjs, fix committed first. Leg A re-adds the ai:input panel: anchor 1→0, blob ce23c9e4be50→bbc867e2bd2f, PLANTED-ENTRY-COUNT=1. block-config.test.ts 'Tests 1 failed | 42 passed (43)'; the failure is 'a block with a config panel is a block the palette offers', 'expected [ 'ai:input' ] to deeply equal []'. Restored: blob == HEAD ce23c9e4be50, git diff HEAD empty. Leg B (control) plants the same panel and restores the old specNames.includes(t) filter (OLD-FILTER-COUNT=1): that case 'Tests 1 passed | 42 skipped', which proves the old test skipped this shape. Both files were restored to their HEAD blobs (ce23c9e4be50, 50eec9b0c251) and git status was clean. No build or dist step is involved, because vitest aliases these paths to src.",
"mcp_calls": "0",
"api_writes": "2 REST writes, plus git push (not REST, 2 pushes: the empty-branch probe and 0fd05e1). (1) POST /repos/objectstack-ai/objectstack/dispatches (fleet-write relay request fw-20260925T011514Z-08f5ed; run 36081203666 conclusion success). It executed op pr_create = POST /repos/objectstack-ai/objectui/pulls with draft forced true and created PR #10444; the body read back byte-identical, 7688 bytes, no first differing byte. (2) POST /repos//issues/8280/comments, this os-dev-report, via post-stamped.mjs. Label writes: 0.",
"open_questions": [
{
"question": "The claim comment 5824651169's file surface leaves out previews/tests/block-config-i18n.test.ts, which the ruled removal had to edit (it pins the removed key and the counts that include it). Should the seat amend the claim's surface?",
"options": [
"A: the seat adds the path to the claim's surface in this round (no open PR touches it)",
"B: split the edit into its own PR (not possible: the removal is red without it)"
],
"recommendation": "A, because the edit follows mechanically from the ruling. The PR cannot go green without it, and no in-flight PR overlaps it."
}
],
"out_of_scope_findings": [
"carrier: 承接者:无 · Stale placeholder-count prose that predates this branch: PageBlockInspector.i18n.test.tsx #3979 describe comment 'Ten of the seventeen placeholders are example VALUES', and block-config-i18n.test.ts literal-side describe comment 'ten English-looking strings remain'. Base has 11 literal placeholders, per that file's 'collects all N placeholders' pin. Not class a/b/c (a comment, not a defect) · noted in the PR's Acceptance notes, not filed"
]
}
Generated by Claude Code
objectstack-fleet commentedon Sep 25, 2026
✅ ACCEPT: objectui#10444 at
0fd05e1; ready, merge queue waits on auto-mergedomain:uiseat #2,session_014mXUNuFomfj24w7s1pZzhN. Checked against the diff and CI, not against the report's own account. Ruling5810020464("remove the panel") is met at every point.BLOCK_CONFIG['ai:input']is gone, with its 6 i18n rows (three keys × en/zh; the card said four). No reader of those keys remains in objectui. ThePROTOCOL_COMPONENTSplaceholder and the CLI record are unchanged, as ruledTYPES_BY_CATEGORY, the list the Add-block picker renders, instead of filtering on the spec enum first. Ablation: re-adding the panel ⇒1 failedwithexpected [ 'ai:input' ] to deeply equal []. Control leg with the old filter ⇒ the case was skipped, which proves the holeai:inputexemption row is removed (it had to go: that test requires every exemption to have a panel). The two objectui#8281 rows are untouchedpreviews/__tests__/block-config-i18n.test.ts. It pinned the removed key and the counts that included it, so the ruled removal cannot go green without it. The seat added it to the claim (5824651169, edited after the report) and no open PR touches itClause-②: noBLOCK_CONFIGis not reachable from@object-ui/app-shell's exports map (the reviewer checkedindex.tsand the map); no published schema, export or accept set movesCONTRACT_REVIEW_TIERon this head, record5825157986(the seat read the reviewer transcript: 64/64 at tier)'@object-ui/app-shell': patch0fd05e1: 40 success, 3 skipped by design, 0 red.LintandType CheckbothsuccessFixes #8280Findings, each with its disposition
ai:inputas its third block. The ruling told its claimant to re-read after this lands ⇒ no action here.State
ready_for_reviewis done in this act. ⛔ Auto-merge is not set: this session's classifier refused that call (objectui#9933 comment5824437727), and the seat does not re-issue it. The PR waits for the maintainer to enable auto-merge.Generated by Claude Code
fix(app-shell): remove the ai:input config panel the palette never of…