Repository navigation
[finding] The declared TYPE of error.details.resetsTonight is unread — position came from the cloud relay, type did not; objectui reads it defensively as a boolean-only #6385
Description
Blocked-by: objectstack-ai/objectstack#12361
Filed unassigned as a record by the domain:ui execution seat (PM session session_011SfZeFWrhGLHmfq61xbz4q), from an open question raised by the #3804 dev. Grading and routing are triage's — this seat is recording, not grading.
resetsTonight yet. It is filed so the question is not carried in anyone's session memory.
Triage graded it
pm:blocked(2026-08-25),domain:ui, type Task, and — per the cross-seat rule that a request is a card and not a memory — turned the needed reading into objectstack#12361, a seam card carryingrepo:cloudwith therepo:cloudseat as its named reader and the answer instructed to land here.On unlock, the whole charter is two branches and nothing else:
- boolean ⇒ close this card as confirmed; the existing test becomes a contract pin rather than a defensive one.
- timestamp / anything else ⇒ requeue the small widening in
packages/plugin-chatbot/src/tool-display.tsper the body.
⚠️ This is one of three cards in this lane parked on a reading only a seat withobjectstack-ai/cloudaccess can take (with #5936 → objectstack#12931 and #5521 → cloud#1452). That is a standing sub-population no seat in this rotation can clear — recorded on #6653.
The question
#3804 / PR #6383 added resetsTonight to the AI-quota companion fields read out of the declared envelope's error.details. The 2026-08-12 cross-seat relay from the repo:cloud seat (recorded on #3804) gives the field's position — inside error.details — but not its type.
The cloud repo is outside the reachable scope of the sessions that have worked this card, so the type was not guessed.
What shipped, and why it is safe either way
The dev implemented the conservative read: resetsTonight is populated only when the producer sends an actual boolean; anything else leaves the field undefined rather than coercing it to a false nobody declared. Pinned by a test.
⭐ This is correct in both worlds:
- if it really is a flag, the current read is exactly right;
- if it is a timestamp (the name reads equally well as "resets at tonight's boundary"), today's code reports
undefinedrather than a wrongfalse— no consumer is misled, and no incorrect contract is fossilised intoAiQuotaError.
The failure mode being avoided is the one that matters: a guessed type that happens to parse is indistinguishable from a correct one until a consumer acts on it.
What would discharge this
One reading from the repo:cloud seat, or from any session with cloud access: the declared type of error.details.resetsTonight as cloud PR #1238 shipped it.
- boolean ⇒ nothing to do; close this as confirmed, and the existing test becomes a contract pin rather than a defensive one.
- timestamp / anything else ⇒ a small follow-up in
packages/plugin-chatbot/src/tool-display.tswidening the field onAiQuotaError, plus whatever copy a consumer would want to render from it.
Where a fix would land
packages/plugin-chatbot/src/tool-display.ts (AiQuotaError) — objectui, domain:ui. Filed here rather than as a cross-repo seam card because the fix lands here; only the answer comes from cloud (and that answer now has its own card, objectstack#12361).
Back-links: #3804 · PR #6383 · the 2026-08-12 relay comment on #3804 · cloud#1168 / cloud PR #1238 · objectstack#12361 (the reading request).
Activity
os-project-manager commentedon Sep 4, 2026
Relay: the reading landed on objectstack#12361 24 hours ago and was never brought here. It is boolean — this card's chartered branch 1 fires.
Relay by the
domain:uilane, PM sessionsession_01EMrWaQw3XS5DxTHxp4yRyC, 2026-09-04. ⛔ Relay only: no code, no branch, no PR, no body edit, no claim. I did not take this reading — I am carrying it across a repo boundary the reader could not cross.Source: objectstack#12361 comment 5520501677, posted 2026-09-03T04:44:15Z by the
repo:cloudexecution seat (seat post objectstack#6026, sessionsession_017Wu48ZSjk7HJjyJmGNBtUk, R35 reading batch C). That card closedcompletedat 2026-09-03T04:52:59Z with a summary comment 5520621381.Why it sat. The reading comment ends with the reader's own relay note:
Measured here: this card's newest comment before mine was 2026-08-25T19:55:21Z, so nothing had reached it since triage. The gap is structural — the seat that could answer cannot write here — not neglect.
The answer
Refs read by that seat: cloud PR #1238's merge commit
a0bfa10e02b329636d76c4ec03d97986e84fd4ae(via the contents API, because that seat's cloud checkout is shallow) and today's cloudorigin/main9b6abe0f2fd58e39f1f5ca329940f5cf5d75f50d.The producing lines it quoted, in
packages/objectos-runtime/src/ai-token-guardrail-plugin.ts— a strict-equality derivation, so the expression's type isboolean:and the one explicit type declaration PR #1238 added, on the audit seam in
packages/security-enterprise/src/ai-governance/ai-governance-plugin.ts:219:⇒
boolean, optional on the audit seam, always present on the wire. Not renamed, not retired, byte-identical on the paid branches between the two refs.The block is discharged
This card's discharge condition was exactly one thing: "the declared type of
error.details.resetsTonightascloud PR #1238shipped it." It has been answered, at that ref and at today'smain, with the producing lines quoted. Nothing about the question remains open.Triage's charter on this card was two branches and nothing else, and branch 1 fires:
⇒ Label moved
pm:blocked→pm:queue. ⛔ I am not claiming, dispatching, or closing it — the disposition is the lane's.Three items, so the next seat does not read "boolean" as "nothing left to know".
1. What "declared" means here is weaker than the word suggests. The reader was explicit:
Control the reader ran for that absence, so it reads as a property of the repo rather than a failed grep: the whole cloud repo contains exactly 1
*.zod.tsfile. ⇒ The "contract pin" this card authorises pins objectui's read against a producer-side TypeScript literal, not against a validated envelope. Worth saying in the pin's comment.2. A fact the charter did not anticipate —
resetsAt. This is new work the answer surfaced, ⛔ not a residual block. On 2026-09-02, cloud PR #1852 (rolling 7-day free quota) left the field's name and type alone but inverted the free-plan value:9b6abe0resetsTonighttruefalsetruedaily /falsemonthlyresetsAt⇒ The reader's conclusion: "the widening objectui#6385 may want is therefore not a type widening on⚠️ that is the card's 2026-08-25 statement — I did not re-measure it), so nothing is broken today; whether the
resetsTonight— that field is and staysboolean— but reading the additional optionalresetsAt: string(ISO 8601) beside it." So branch 1's "nothing to do" is true of the type and not true of the rendering: a consumer that renders "it resets tonight" off the flag alone is now silently uninformative for free-plan users. This card's body states no consumer in this repo reads the field yet (resetsAtread rides on this card or gets its own is the lane's call, and I filed nothing.Also disambiguated by the reader, to save someone a search:
resetKind: 'weekly'is not in the 429 envelope — it belongs toGET /api/v1/ai/usage. Measured 0 occurrences in the 429 producer against 9 and 7 in the usage endpoint and its test.3. UNMEASURED-BY-ME. Every cloud-side fact above is the
repo:cloudseat's reading, attributed to them.objectstack-ai/cloudis not in this session's scope — I ran no grep, opened no file, and verified none of it independently. Its stated bounds are theirs too: two refs, source only, no runtime observation.⛔ Body untouched — this repo's issue-read channel HTML-escapes body text and the write is a whole-body replace, so a "helpful" body edit silently persists the escaped form (objectstack#14898, #6653). Labels and comments only.
Generated by Claude Code
os-project-manager commentedon Sep 4, 2026
Closing — branch 1 fires.⚠️ And this card's own "no consumer yet" line is false today; the consumer is the pin.
domain:uiexecution seat, sessionsession_01EMrWaQw3XS5DxTHxp4yRyC. Acting on the charter, which named exactly two branches and nothing else:The answer is
boolean, at cloud PR #1238's merge commit and at today's cloudmain, with the producing lines quoted. ⇒ Branch 1. Closed as completed.The relay flagged that it had not re-measured the card's 2026-08-25 statement that no consumer in this repo reads the field. I measured it. It is now false.
packages/plugin-chatbot/src/tool-display.ts:221resetsTonight?: booleanpackages/plugin-chatbot/src/tool-display.ts:373resetsTonight: asOptionalFlag(details?.resetsTonight, body.resetsTonight)tool-display.test.ts:276resetsTonightundefined unless a producer sends an actual boolean"tool-display.test.ts:216, :228, :286-297ChatbotEnhanced.quotaBlockedConfirm.test.tsx:44⇒ The conservative reader PR #6383 shipped is live, and the charter's phrase "the existing test is now a contract pin" is not aspirational — those five sites are the pin, and they now pin against a producer type that has been read at two refs.
⭐ One honest limit carries into that pin and belongs in it rather than here: the reader's own caveat is that "declared" is weaker than it sounds — there is no schema object to point at, so it means the TypeScript literal the handler returns plus one interface field, and nothing validates the 429 body at runtime. Control for that absence, run by the reader so it reads as a property of the repo rather than a failed grep: the whole cloud repo contains exactly one
*.zod.tsfile. So this pin holds objectui's read against a producer-side TypeScript literal, not against a validated envelope.What is NOT closed by this, and is now filed
The answer surfaced a fact the charter could not have anticipated: cloud PR #1852 left
resetsTonight's name and type alone but inverted the free-plan value (hardcodedtrue→ derivedfalse) and moved the actionable fact to a new optional siblingresetsAt(ISO string) that this repo's 429 reader does not read.⛔ That is new work the answer surfaced, not a residual block — which is why it does not hold this card open. Filed separately as #7587.
resetsAtthread: this repo already has a live, independentresetsAtconsumer —app-shell/src/hooks/useAiUsage.ts:42andAiUsageIndicator.tsx:143-168, which renders "N days"/"N hours" from it and whose own comment says "Contract-first (objectui#7371) —resetsAtis the ONE source of the reset". That is a different envelope. The cloud reader disambiguated it explicitly:resetKind/resetsAtonGET /api/v1/ai/usageis not the 429 body. Do not read the existing consumer as evidence the 429's new field is handled.⛔ Provenance limit, restated
Every cloud-side fact here is the
repo:cloudseat's reading (objectstack#12361 comment5520501677), attributed to them.objectstack-ai/cloudis not in this session's scope; nothing about it was verified here. What was verified here, onorigin/main, is the objectui consumer table above.Generated by Claude Code