finding(plugin-chatbot): the 429 quota envelope gained an optional `resetsAt`, and free-plan `resetsTonight` flipped `true` → `false` — `tool-display.ts` reads only the flag, so free-plan users now get a reset promise with nothing behind it

[os-project-manager]

Filed by the domain:ui execution-seat PM (session session_01EMrWaQw3XS5DxTHxp4yRyC), out of #6385's discharge. That card asked one question — the declared type of error.details.resetsTonight — got its answer (boolean, confirmed at two cloud refs) and closed on branch 1. ⛔ This is new work the answer surfaced, not a residual block on it.

The producer moved under the consumer

The repo:cloud seat's reading on objectstack#12361 (comment 5520501677, 2026-09-03) records that cloud PR #1852 — rolling 7-day free quota, landed 2026-09-02 — left the field's name and type alone and changed what it means:

	free plan, before	free plan, on cloud 9b6abe0	paid plan, both refs
resetsTonight	hardcoded true	derived ⇒ false	true daily / false monthly
resetsAt	absent	present, ISO 8601 string	absent

In the reader's own words: "the widening objectui#6385 may want is therefore not a type widening on resetsTonight — that field is and stays boolean — but reading the additional optional resetsAt: string (ISO 8601) beside it."

What this repo does today — measured on origin/main ccb3ad78a

packages/plugin-chatbot/src/tool-display.ts:

:221   resetsTonight?: boolean;
:373   resetsTonight: asOptionalFlag(details?.resetsTonight, body.resetsTonight),

Repo-wide, resetsAt appears nowhere in plugin-chatbot. ⇒ The 429 reader takes the flag and nothing else.

⚠️ The consequence is not a crash — it is a silence. A free-plan user who exhausts the allowance now gets resetsTonight: false where they used to get true, and the one piece of information that replaced it (resetsAt) is dropped on the floor. Whatever the UI said about when the quota returns, it now says less, for the plan that most needs to know.

⚠️ Two things this card does NOT claim

1. That a user-visible string is currently wrong. tool-display.ts populates a field; whether and how any surface renders it is a further question this card has not measured. Someone taking this should establish the render path before choosing the repair — the difference between "renders nothing" and "renders a false promise" changes the priority.

2. That the existing resetsAt consumer is relevant. ⭐ This repo already reads a field of that name — app-shell/src/hooks/useAiUsage.ts:42 (resetsAt: string | null) and AiUsageIndicator.tsx:143-168, which renders "N days" / "N hours" and whose own comment reads "Contract-first (objectui#7371) — resetsAt is the ONE source of the reset".

⛔ That is a different envelope. The cloud reader disambiguated it explicitly: resetKind / resetsAt belong to GET /api/v1/ai/usage, measured at 0 occurrences in the 429 producer against 9 and 7 in the usage endpoint and its test. Reading the existing consumer as evidence the 429's new field is handled is the trap this section exists to prevent.

The pin that already exists, and why it does not cover this

#6385 closes with tool-display.ts's reader promoted to a contract pin — five sites, including "leaves resetsTonight undefined unless a producer sends an actual boolean" (tool-display.test.ts:276). That pin is about the field's type, and the type did not move. A value inversion and a new sibling key are exactly what a type pin cannot see.

Provenance and limits

Every cloud-side fact above is the repo:cloud seat's reading, attributed to them; objectstack-ai/cloud is not in this session's scope and nothing about it was verified here. The objectui-side measurements — the two tool-display.ts sites, the absence of resetsAt in plugin-chatbot, and the separate useAiUsage / AiUsageIndicator consumer — were taken here on origin/main.

⛔ Not claimed, not graded — lane and priority are the triage seat's.

Related: #6385 (the discharged question this came out of) · objectstack#12361 (the cloud reading) · #7371 (the contract-first resetsAt rule on the usage surface) · PR #6383 (the conservative reader now serving as the pin).

[os-justin]

Claim: PM loop round R2
Session: session_01YBWFb5YgMU5dw8p2VKj16S
Branch: claude/issue-7587-429-resetsat-free-plan
Worktree: objectui-issue-7587
Domain: domain:ui
File surface: packages/plugin-chatbot/src/** (tool-display.ts, its tests, and whatever renders the 429 envelope inside this package) + a changeset. ⛔ packages/app-shell/** is OUT — both because useAiUsage / AiUsageIndicator are a different envelope (the card's §2 trap) and because #7234 holds that package this round (stop on breach; explain in the report)
Container & model: M, mode:subagent, model: TIER_DEFAULT (opus) — ⚠️ scripts/pm/dispatch-gates.mjs does not exist in objectui, so no --tier output is quotable; gate family derived by hand at 17:2xZ. Tier read from objectstack:scripts/pm/dispatch-gates.mjs:10023: sonnet = FLOOR, opus = DEFAULT, fable = CONTRACT_REVIEW_TIER (measured unavailable, rate_limit HTTP 429), so the exemption caps this at opus — also the right tier on its own merits, since the repair choice depends on a render-path measurement
Clause-②: no — reading an additional optional field the producer already sends, on the consumer side, changes no schema's accept set and widens no authorable surface. ⚠️ If the implementation ends up declaring or exporting anything new, that flips it — stop and report rather than proceeding
Comment thread read: 0 comments — verified by an explicit get_comments call returning [], ⛔ not inferred from a missing count field. ⚠️ Note for triage: this card carries bug / pm:queue / finding / priority:p2 / domain:ui but no audit comment, so the grading rationale is not on the card
Serial constraints cleared: all five R1 PRs MERGED; packages/plugin-chatbot was last touched by #8077 (d3499b315, the chatbot-floating spread fence, landed today ~15:26Z) — ⚠️ same-day churn in the same package, re-read renderer.tsx and the __tests__ dir on current main before assuming anything about them. ⛔ In flight this round: #7234 holds packages/app-shell. #7703 / #7704 sit behind this card on plugin-chatbot. No in-flight sibling asserts a pin on this card's behaviour.

⛔ Assignee field only — this says a PM has claimed the card. It does not say the dev's own claim protocol is satisfied; the dev posts its own Claim: with its own session ID and branch.

---

Generated by Claude Code

[os-justin]

Claim: dev seat (R2)
Session: session_01YBWFb5YgMU5dw8p2VKj16S
Branch: claude/issue-7587-429-resetsat-free-plan (pushed empty at cf1d29e2f)
Worktree: objectui-issue-7587
File surface: packages/plugin-chatbot/src/** + a changeset. packages/app-shell/** is OUT (different envelope; held by #7234 this round).
Assignee field untouched.

Card body read in full on GitHub; not sanitizer-truncated (no angle-bracket loss, all sections and the closing Related: line intact). Comment thread at claim time: one comment, the PM's dispatch claim — no triage/audit rationale on the card.

Re-measured on origin/main cf1d29e2f (the card measured ccb3ad78a): tool-display.ts:221 resetsTonight?: boolean and :373 asOptionalFlag(...) — both line numbers still hold; resetsAt still 0 occurrences repo-wide in plugin-chatbot.

---

Generated by Claude Code

[os-justin]

os-dev-report

{
  "issue": 7587,
  "status": "done",
  "branch": "claude/issue-7587-429-resetsat-free-plan",
  "base": "cf1d29e2f",
  "head": "8fc2ab45d",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8131",
  "pr_state": "draft (left draft; no ready-flip, no enqueue, no auto-merge)",
  "premise_still_valid": false,
  "premise_detail": "Every objectui-side measurement in the card re-verified EXACT on cf1d29e2f: tool-display.ts:221 `resetsTonight?: boolean`, :373 `asOptionalFlag(...)`, and `resetsAt` at 0 occurrences repo-wide in plugin-chatbot. What is falsified is the card's TITLE claim -- 'free-plan users now get a reset promise with nothing behind it'. Measured render path: `resetsTonight` has zero readers after the parse (useAiQuotaCopy uses message/messageEn/topUp; the two other parseAiQuotaError call sites are yes/no predicates), and parseAiQuotaError/AiQuotaError are not in the package barrel, so there is no external consumer either. The banner text is exactly 'Upgrade needed' + the server's own messageEn + 'Upgrade plan'. The answer to the card's own unmeasured question is RENDERS NOTHING, not RENDERS A FALSE PROMISE -- no user-visible string in this repo is derived from the flag, and the true/false flip changes not one rendered byte. Per dispatch: the grade should drop.",
  "cloud_premise": "unverified from here. objectstack-ai/cloud is not in this session's repo scope; 'free plan now sends resetsTonight:false plus resetsAt' remains the repo:cloud seat's reported reading, attributed as such and not confirmed. In-repo corroboration exists but is indirect and is NOT evidence about the 429 producer: packages/app-shell/src/layout/AiUsageIndicator.tsx:142 names cloud PR #1852 and renders resetKind 'weekly' from the USAGE endpoint's resetsAt. Also unverifiable and load-bearing for any renderer: the 429's wire POSITION of resetsAt (error.details vs top-level) is an inference from the cloud seat's prose, never measured. A reader built on a guessed position fails silently.",
  "files_changed": [
    "packages/plugin-chatbot/src/__tests__/ChatbotEnhanced.quotaResetPromise-7587.test.tsx (new, 127 lines)",
    "packages/plugin-chatbot/src/tool-display.test.ts (+38: free-plan whole-envelope value pin + `'resetsAt' in parsed` is false)",
    "packages/plugin-chatbot/src/tool-display.ts (+18, comment only: the measurement recorded at the field)",
    ".changeset/7587-quota-reset-promise-pin.md (new, empty frontmatter)"
  ],
  "diff_stat": "4 files changed, 193 insertions(+), 0 deletions(-)",
  "line_budget": "n/a -- diff touches no skills/** file, so no skills line ratchet applies. Net +193/-0, all of it test + comment + changeset.",
  "gates": {
    "pnpm exec vitest run packages/plugin-chatbot/": "exit 0 -- Test Files 39 passed (39), Tests 461 passed (461); re-run clean after all ablations",
    "pnpm --filter @object-ui/plugin-chatbot type-check": "exit 0 (tsc --noEmit && tsc -p tsconfig.test.json). First run exited 2 with TS2307 on @object-ui/components et al; cause was an unbuilt dependency closure, cleared by `pnpm --filter '@object-ui/plugin-chatbot^...' build` (exit 0), not by any source change",
    "check:control-bytes": "exit 0 -- scanned 6499 tracked text files",
    "check:published-dist": "exit 0 -- 39 packages, 0 tooling artifacts in build output",
    "check:published-tsconfig-exclude": "exit 0 -- 34 enforced packages carry the directory form",
    "check:handler-key-reads": "exit 0 -- 62 reads judged, 0 unjudged",
    "check:sdui-registration-pins": "exit 0 -- 16 registrations present in the built console",
    "node scripts/check-changeset-presence.mjs": "exit 0 -- 2 source files of 1 released package changed, 1 changeset declared, empty frontmatter accepted as the explicit exemption",
    "lint": "pnpm --filter @object-ui/plugin-chatbot lint exit 0 -- 90 problems, 0 errors, all warnings pre-existing; zero findings in the three changed files (the one tool-display.ts warning is :346 `let body: any`, untouched by this diff)",
    "ablation_1_flag_driven_reset_copy": "vitest exit 1 -- invariance pin RED, accounting pin green. Mutation proven on disk (marker 0 -> 1, original block 1 -> 0); restore proven by empty `git diff HEAD` and marker back to 0",
    "ablation_2_unconditional_reset_copy": "vitest exit 1 -- accounting pin RED, invariance pin green. Same on-disk and restore proofs",
    "ablation_3_parser_reads_resetsAt": "vitest exit 1 -- parser pin RED, 1 failed / 49 passed. Pre-mutation blob hash verified equal to HEAD's; restore proven by blob hash back to ab7877ff22ae (== HEAD) and empty diff"
  },
  "deviations": [
    "Repo-wide `pnpm lint` (turbo run lint) NOT run; narrowed to `pnpm --filter @object-ui/plugin-chatbot lint`, which is the same eslint invocation turbo runs for this package. Narrowing evidence: all three changed source files live in that one package; the diff changes no eslint config, so no untouched file's verdict can move; object-ui/no-unused-imports (the error-severity rule the dispatch named) reported 0 errors. Farm-wide lint is left to CI.",
    "vitest run scoped to packages/plugin-chatbot/ rather than the full `pnpm test`, per the repo's path-filter convention; a sibling dev is in flight in this container.",
    "No test process was killed by name; no pkill was used at any point.",
    "One MCP channel switch: dedup for the out-of-scope finding used a targeted `search_issues` (2 calls: the finding's terms plus a control). The control fired -- a query naming resetsTonight/resetsAt returned #7587 itself -- so the empty result for the finding's own terms is a real reading, not a dead channel."
  ],
  "mcp_calls": 9,
  "open_questions": [
    {
      "question": "resetsTonight is now a declared, populated, twice-pinned field with ZERO readers in this repo and no place in the package barrel. Wire it, or retire it?",
      "options": [
        "A. Leave as-is (this PR): field parsed and unread, both the inertness and the parse's silence pinned, decision deferred with the reasoning written at the field.",
        "B. Wire a renderer: read resetsAt in parseAiQuotaError and render a reset time in the quota banner. Needs two things this repo cannot supply -- a MEASURED wire position for resetsAt in the 429 body (error.details vs top-level) from the cloud seat, and a ruling that a second reset sentence beside the server's own prose is wanted, which objectui#7253 deliberately ruled against ('one 429 can never be described by two different sentences on the same screen').",
        "C. Retire resetsTonight from AiQuotaError entirely as declared-but-unenforced, deleting the #6385 type pin with it."
      ],
      "recommendation": "A now, and card B behind a cloud-side measurement request rather than an objectui card -- B's blocking input is a producer fact, so an objectui card for it would sit unactionable. C is defensible on dead-code grounds but should not be taken while B is open: retiring the flag and then wanting reset copy would mean re-deriving the same envelope twice."
    },
    {
      "question": "Does the free-plan user learn when the allowance returns at all, today?",
      "options": [
        "Yes, but on a different surface and a different envelope: AiUsageIndicator (app-shell) renders 'Resets in N days / N hours' from the USAGE endpoint's resetKind 'weekly' + resetsAt.",
        "No, if the user never opens that indicator; the chat banner says only what the server's prose says."
      ],
      "recommendation": "Treat as a product question for the triage seat, not a defect. Recorded here because it is what makes the measured severity low: the information exists in the product, just not on this surface, and the chat banner's sentence is server-owned so cloud can change it without touching this repo. Explicitly NOT offered as evidence that the 429's resetsAt is handled -- it is not (the card's section-2 trap)."
    }
  ],
  "out_of_scope_findings": [
    "filed as #8130: two files named tool-display.test.ts in packages/plugin-chatbot (460-line quota-pin file outside __tests__/, 91-line helper file inside it). Repo-wide there are exactly 2 such (package, basename) collisions. A path-filtered run of the __tests__ one exits 0 having run none of the quota pins, and objectui#3240's zero-match protection cannot fire because both spellings match a real file. Filed unassigned, label `finding`, no lane/priority asserted."
  ],
  "notes": [
    "Card body read in full on GitHub and checked for sanitizer truncation: intact (no angle-bracket loss, all sections present through the closing `Related:` line).",
    "Comment thread at claim time held exactly one comment, the PM's own dispatch claim -- consistent with the dispatch's `0 comments` reading, which was taken before that claim was posted. No triage/audit rationale exists on the card.",
    "Same-day churn re-read as instructed: PR #8077 (d3499b315) touched renderer.tsx and its __tests__; this diff touches neither file, and the full package suite is green over both.",
    "Clause-②: unchanged, still `no`. The implementation declares and exports nothing new -- the production diff is comment-only; no interface field, symbol, or barrel entry was added. The one new file is a test.",
    "Assignee field untouched. PR left draft. No labels applied to #7587. No status on any PR was flipped."
  ]
}

---

Generated by Claude Code