[Epic] v17 bug focus — Seat A: platform bugs (objectstack)

[hotlong]

Theme charter for the post-v17 priority regime. Filed by the (former) skills PM seat, session session_018WuTtyckQa1VcXwgd52JpN, on maintainer instruction.

Authority (maintainer, 2026-08-14, verbatim, untranslated)

v17 版本发布了，我下一步不想再追求每个车道任务清零，而是应该按照任务优先级选择重点任务先开发。

所以我现在希望集中处理 bug，应该怎么安排。

先只考虑 objectstack 和 objectui 仓库

可以继续用 target:v17

把两张主题卡（epic）开出来

Scope

All open bug-labeled cards in objectstack (37 at filing), every domain — EXCEPT cards already in flight under the two live lane seats: domain:devx (#6023) and domain:spec (#6017). Console-facing cards (repo:objectui) belong to the sibling theme, Seat B (see the epic filed immediately after this one; the two cards cross-reference in the comments).

Workability rule (the default flip): a card is workable under this theme only when it carries target:v17. Everything else stays frozen stock — no lane-clearing. The PM nominates second-batch candidates with a short list on this epic; the maintainer approves by labeling (or one-word ruling), and only then do they enter the queue.

First batch (workable now or pending one maintainer ruling)

• [bug] OIDC SSO registration is broken end-to-end: the bridge always sends oidcConfig.mapping.id, which @better-auth/sso@1.7.0-rc.2 rejects as an unrecognized key #8193 — OIDC SSO registration broken end-to-end (pm:blocking)
• The audience-binding-suggestion reconciler writes with a tenant-less system context, so one organization-less row serves every tenant — the second half of the #8577 install-path dead end #8617 — tenant-less reconciler row serves every tenant (security, target:v17)
• reapRotatedSecret deletes the IN-FORCE ciphertext when a SettingsEngine adapter drops context, leaving a dangling value_enc and an unreadable setting #8262 — reapRotatedSecret deletes the in-force ciphertext (target:v17)
• [security] GET /api/v1/meta/datasources still serves stored cleartext credentials — the metadata read path has no per-type redaction hook #8154 — /meta/datasources serves cleartext credentials (security, target:v17)
• [security] sys_account stores live third-party OAuth access/refresh/id tokens as plain columns, and the object is API-readable #7987 — sys_account stores live OAuth tokens in plain columns (security, target:v17)
• [finding] API keys carry no organization — under the isolated posture a minted key reads no org data at all (no leak, but the key surface is inert) #8287 — API keys carry no organization (security, P0) — pending the maintainer's P0 ruling

P0 rulings pending on the maintainer's desk (fix-in-17.x / defer-v18 / void): #8500, #8282, #8096, #8270 (those four are Seat B surface), #8287. Decision-gated bugs pending one-sentence rulings: #8629, #8586, #8421.

Before dispatching anything: orphan re-verification

Former lane PMs stood down without closing out. Verify the true state (PR? merged? labels?) of each and either close out or re-dispatch under this theme:

• domain:identity ledger named sharing rules: the business_unit recipient expands the whole SUBTREE, but the spec declares it as exactly one unit (no subtree) #7807, A successful /auth/change-password over a BEARER token never clears must_change_password — the caller is permanently locked out of every protected route with a 200 telling them it worked #8049, [finding] The bulk AST write path passes opCtx.operation raw, so purge / transfer / restore derive no row scope while update / delete now do (#7665 asymmetry) #7809 as dev-in-flight (2026-08-12) — none appear in the current target:v17 listing, so they may have landed or lost their labels
• domain:drivers ledger named finding: after the #7598 Q1=B ruling, a read scope with a driver-refused field reference answers 400 from the driver — which cuts across #5367's attribution argument on that one path #7929 as in flight
• domain:cli ledger named 【缺陷】数据导出(CSV/XLSX)日期时间列硬编码按 UTC 渲染,与界面时区不一致(@objectstack/rest export-format.ts formatDate) #8373 (draft PR open) and RuntimeConfigPlugin's features.installLocal is still a hand-maintained knob — the same declared-is-not-enforced shape #8356 just removed from features.marketplace #8388 (ACCEPT awaiting landing)

Operating rules

• WIP cap: ≤3 dev agents in flight.
• Pull order inside the theme: priority:p0 > pm:blocking > security > age.
• Bug-fix PRs ride the normal merge queue (auto-merge allowed). Exception: any diff touching .claude/skills/**, skills/**, .claude/agents/os-dev.md, or AGENTS.md is ADR-class — draft + human merge, per standing rules.
• Standard discipline unchanged: claim-before-code with session + branch, per-task worktrees, os-dev dispatch protocol, model tiering (default opus; fable where the mandatory clauses apply), union-after-final-commit with quoted HEAD.
• Reporting: round reports on this epic (not per-lane); decisions surfaced to the maintainer via needs-user-decision or the round report.
• Lane seat posts stay ⏳ vacant; the PM session taking this theme registers here with a claim comment (session ID + scope) and de-registers on stand-down.

Out of scope (by ruling)

hotcrm and cloud repos. Boundary note for the maintainer: hotcrm's four P0s (e.g. #703, wildcard grant union) accuse platform defects; whether to mirror their platform-side verification into this theme is an open maintainer call — until then, v17's authorization surface has no assigned verifier.

Done means

Every in-scope target:v17 bug is closed (fixed, or ruled defer/void with the ruling recorded); the five P0s all carry an explicit maintainer disposition.

[hotlong]

Sibling epic filed: #8668 — Seat B: console bugs (objectui repo + repo:objectui cards here). The two seats partition the in-scope bug set: console-facing surface is B's, everything else objectstack-side is A's.

---

Generated by Claude Code

[qq9340100]

Seat A registration — theme PM session session_01MX1qcBzfwZb5wkRrJTNbhH.

Scope taken as chartered: open bug-class cards in objectstack-ai/objectstack carrying target:v17, excluding the repo:objectui surface (Seat B, #8668) and cards live under domain:devx (#6023) / domain:spec (#6017). WIP cap ≤3. De-registers on stand-down.

1. Orphan re-verification — complete, nothing to re-dispatch

The charter asked for the true state of six cards named in stood-down lane ledgers. All six are closed with a MERGED PR; none needs re-dispatch or close-out work:

card	state	landed by
#7807	closed / completed	PR #8097 MERGED
#8049	closed / completed	PR #8101 MERGED
#7809	closed / completed	PR #8106 MERGED
#7929	closed / completed	PR #8198 MERGED
#8373	closed / completed	PR #8487 MERGED
#8388	closed / completed	PR #8466 MERGED

Cosmetic residue only: #7807, #8049, #7809, #7929 still carry pm:dispatched on a closed card. Harmless (no view reads pm:* on closed issues) and left alone rather than mass-edited.

2. Workable inventory under the charter's target:v17 flip

target:v17 ∩ open ∩ objectstack-side ∩ not repo:objectui = six cards. Dispositions:

card	lane	state	this round
#7987	metadata	pm:queue, security	dispatched
#8154	metadata	pm:queue, security	dispatched
#8617	identity	pm:queue, security	dispatched
#8149	services	pm:queue, security	held — batch collision, see below
#8262	services	pm:queue	held — round 2
#8337	drivers	pm:queue	routing conflict, needs triage — see below

First-batch cards from the charter that are not dispatchable: #8193 is pm:blocked and already assigned (not touched, per the never-touch-another-claim rule); #8287 is pm:blocked with no pm:queue, awaiting the maintainer's P0 ruling.

Pull order applied as chartered (priority:p0 > pm:blocking > security > age): no workable card carries p0 or pm:blocking, so the three oldest security cards lead. That would be #7987, #8149, #8154 by age — #8149 was displaced by #8154 on batch independence, not on rank: #8149's unlock re-check resolves it into a consumer application of the internal: true + batch privileged accessor that PR #8348 landed in packages/objectql, which is the same mechanism and plausibly the same file #7987 touches. Recording the collision here rather than leaving it to be rediscovered next round. #8149 leads round 2.

3. Gate discharges verified this round (not inherited from the cards' own text)

• [security] GET /api/v1/meta/datasources still serves stored cleartext credentials — the metadata read path has no per-type redaction hook #8154 — both gates discharged: A per-type metadata redaction seam belongs in @objectstack/spec/kernel — no service package can reach a registry in metadata-protocol #8300 closed via PR feat(spec): per-type metadata redaction seam in kernel, one credential-key definition in data (#8300) #8530 MERGED (registerMetadataTypeRedactor present at packages/spec/src/kernel/metadata-type-redaction.ts, packages/spec/src/data/datasource-credential-redaction.ts present on origin/main), and [finding] metadata-protocol interpolates raw driver text into client-facing messages — three downstream sanitizers each have a hole because of it (option C of #8086) #8136 closed. Reverse-checked that no consumer of the seam exists outside packages/spec, so the card is still fully open rather than partly delivered.
• [security] sys_account stores live third-party OAuth access/refresh/id tokens as plain columns, and the object is API-readable #7987 — blocker Check whether sys_session.token — a live session credential — serializes over the data API (ADR-0100 channel 3 has no read protection) #7823 closed via PR fix(platform-objects): sys_session.token stops serializing on the data API — internal: true (#7823) #7996 MERGED. Premise re-measured on origin/main 189a732: access_token / refresh_token / id_token are still plain Field.textarea at sys-account.object.ts:181-194 with apiEnabled: true, apiMethods: ['get','list'] at :241-244.
• The audience-binding-suggestion reconciler writes with a tenant-less system context, so one organization-less row serves every tenant — the second half of the #8577 install-path dead end #8617 — blocker Two MORE clean inheritors of the #8323 class the #8554 sweep missed: sys_notification_subscription and sys_audience_binding_suggestion — both measured live #8577 closed. Premise re-measured: const SYSTEM_CTX = { isSystem: true } still at suggested-audience-bindings.ts:43, threaded through seven call sites; the record-not-endorse describe block the fix must delete is still at suggested-audience-bindings-install-path.test.ts:361.

4. ⚠️ #8337 — a routing conflict this seat will not resolve unilaterally

The card is labelled domain:drivers + pm:queue, but the drivers seat measured (and re-measured after #8341 merged) that packages/drivers contributes nothing to the fix — the write-path refusal lands in packages/spec/src/data/driver/common.zod.ts and the read-path redaction in packages/spec/src/data/datasource-credential-redaction.ts / service-datasource. It asked triage four times for a ruling and got none.

Two binding rules make this not mine to dispatch: anything touching packages/spec routes to the domain:spec seat (#6017) regardless of who needs it, and a card narrowing the accept set is domain:spec by the mechanical boundary test — which also makes claude-fable-5 mandatory rather than the suggested opus. Under this theme's own exclusion of domain:spec cards, #8337 is out of Seat A's scope as it actually lands.

Asked of triage / the maintainer: relabel #8337 domain:spec, or rule that Seat A proceeds into packages/spec on it. Until then it is held, not forgotten. Its connect-half measurement (does @libsql/client honour ?authToken= from an authored URL) remains genuinely unmeasured — the drivers seat has offered to take that slice if carved out separately, and it determines severity, not the fix shape.

---

Generated by Claude Code

[qq9340100]

Correction and two additions to the registration comment above (5292576801). PM session session_01MX1qcBzfwZb5wkRrJTNbhH.

Correction: the workable inventory is seven cards, not six

I omitted #7644 (domain:cli, pm:queue, target:v17) from the table. It is a release-verification card, not a code repair, which is why it did not read as bug-lane work on the first pass — but it is on the v17 board and belongs in the count.

Worked it this round rather than just listing it, and it shrank: its residual ask 1 is now measured and satisfied — create-objectstack@17.0.0 is npm latest (16.1.0 no longer is, so the card's user-facing premise has expired) and the shipped package/dist/index.js carries the findStaleNamespacePrefixes self-verify symbol, confirmed by unpacking the real published tarball. Ask 2 (the registry-canary job goes green) is still owed and I did not close it — no auto-filed failure issue is open, but that job is weekly, so silence is equally consistent with "no run since the publish." Full reading on #7644. Remaining scope: one post-17.0.0 workflow-run reading.

Addition 1 — ⚠️ Seat B's board is empty, which the workability rule may not have intended

objectstack-ai/objectui has zero open target:v17 issues. Under the charter's default flip (workable ⟺ target:v17), Seat B's entire workable set is the repo:objectui cards filed here — and that is one card, #7100. #6275 is the only other, and it is pm:on-hold.

The consequence worth a ruling: the four console P0s the charter lists as awaiting your disposition — #8500, #8282, #8270, #8096 — carry no target:v17, so under the default flip they are frozen stock, P0 notwithstanding. Two of them (#8500, #8096) are already pm:dispatched and in flight, which means work is proceeding on cards the workability rule says are frozen. That is a labelling inconsistency, not a judgment call I should make silently: either those P0s should carry target:v17, or the rule should read "target:v17 or priority:p0."

Addition 2 — an untriaged P0 sitting outside both seats

#8652 (【能力需求】plugin-approvals 审批请求可见性) is open with priority:p0 and nothing else — no pm:*, no domain:*, no type. It is a capability request, so by the standing rules it is human-floor (feature class) and not adjudicable by any seat. Flagging it because a bare P0 is invisible to every lane query and will otherwise sit until someone reads the P0 list by hand. Not in Seat A's bug scope; not claimed or labelled by me.

---

Generated by Claude Code