[security] An analytics query surface evaluates caller-supplied content outside the object- and field-level read admission, so a non-admin member can read data they are refused everywhere else — detail withheld pending maintainer

Filing gate: ① a reproducible defect with a named landing site — the analytics query admission at the service door (packages/services/service-analytics). reach: exception, could leak data. It was measured once by the #21120 dev on a fresh showcase boot: a non-admin member received data that every /data and /analytics/query door refuses them (stored datasource credential material included). The dispatched dev's premise check is the second derivation. Reader: the maintainer, who holds the request. Dispatched directly at the maintainer's request, with priority. Dedupe: the analytics security family. Related open cards: #21080 (the native-SQL path skips engine read middlewares) and #20987 / PR #21117 ($contains membership on the read scope). Closed: #20943 (an authored cube member's raw sql expression), #20933 (inferred relationship paths). None covers this position: content supplied by the caller at query time.

QA-source: #21056 · integration-system.datasource-credential-refusal-matrix · negative[0]

What is known publicly

A query shape on the analytics surface lets a caller supply content that the native-SQL strategy evaluates as written.
The object-level read admission judges only the query's base object, and the field-level admission sees no attributable field in that content, so neither admission applies to what the content reads.
A signed-in non-admin member whose base object is readable to them was served data from objects they are refused on every other door.
This is broader than the stored-metadata-body family ([security] Family closeout: stored metadata bodies (stored datasource credential material included) are still reachable outside the redacting doors at several further positions — detail withheld pending maintainer #21120): it is a cross-object read through the analytics evaluation path.

Detail withheld pending maintainer (RUNNER rule 2). Ruling to file and dispatch with priority, verbatim (Claude Code session session_01MRdbfpy4sQT8bUjmMhxsN7, 2026-10-01): 「立脱敏卡、优先派发（推荐）」.

Acceptance

Caller-supplied content on that surface is either refused at the analytics door, or admitted only for content every admission can judge. This is consistent with ADR-0021's "zero raw expressions" and with the outcome recorded on [Decision] analytics field gate (#20917): an authored cube member whose sql is an expression — keep the stand-down, judge its identifiers, refuse it, or retire expressions #20943.
Pinned on both sides: the non-admin refused with nothing evaluated; legitimate queries (declared fields, authored cubes) still answered.

Generated by Claude Code

Claim: PM loop round 4
Session: session_01MRdbfpy4sQT8bUjmMhxsN7
Account: os-bill
Branch: claude/issue-21177-analytics-caller-content-admission
Worktree: objectstack-issue-21177
Domain: domain:services (expected; dispatched with priority at the maintainer's direction ahead of a triage grade)
File surface: the analytics query admission at the service door (packages/services/service-analytics/src/analytics-service.ts and/or a new sibling module beside the field-read admission) + tests + changeset; stop on breach
Container & model: M, mode:subagent, tier default (security fix; dispatch-gates --tier: no path-derived mandate)
Clause-②: no
Thread-read: none
Serial constraints cleared: PR #21117 (#20987, another seat, in flight) owns read-scope-sql.ts, strategies/native-sql-strategy.ts and contains-membership-sql.ts. Those files are ⛔ out of this card's surface, and the fix belongs at the door, not in the strategy. PR #21144 (#21120, in the merge queue) edits analytics-service.ts: this card edits that file only after #21144 is on origin/main, with no stacked branch.

Dispatched with priority at the maintainer's direction. Ruling, verbatim (Claude Code session session_01MRdbfpy4sQT8bUjmMhxsN7, 2026-10-01): 「立脱敏卡、优先派发（推荐）」. The request stays in-session. ⛔ It does not go into any GitHub text.

Landing: security boundary, human floor. The PR stays draft for the maintainer's review.

Generated by Claude Code

Triage: grade after dispatch — `bug` · `security` · `priority:p0` · `domain:services` · `area:reports`. ⛔ The claim, the assignee and the state are unchanged

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-01T13:51Z. ⛔ Not a claim, ⛔ not a dispatch. ⛔ It carries no detail beyond the card's own sanitized text.

The card was dispatched with priority at the maintainer's direction ahead of a grade. This act adds the grade only.

Why p0: it was measured on a fresh showcase boot. A signed-in non-admin member was served data from objects every other door refuses them, stored credential material included. That is a cross-object read at a public door on a stock configuration.
Routing: domain:services, as the claim expected.
Family: [security] An analytics query path can carry caller-supplied member text into the native statement in a tier the field gate does not judge — detail withheld pending maintainer #21156 (p0 suspect, the same holder) names caller-supplied member text that reaches the native statement in an unjudged tier. The two may share a root. Their claims answer fold-or-serial between themselves, and the measurement on either says whether one fix closes both.

Generated by Claude Code

test(rest): retarget the #16019 driver-fault test off the now-closed …

os-dev-report
{
"issue": 21177,
"status": "done",
"branch": "claude/issue-21177-analytics-caller-content-admission",
"pr": "#21190",
"session": "session_01MRdbfpy4sQT8bUjmMhxsN7",
"premise_still_valid": true,
"summary": "Added a caller-content admission at the analytics service door (new module caller-content-admission.ts beside the field-read admission), wired in queryDataset (before compile) and in callCtx (beside #21144's refusal, ahead of the field gate). A member a caller supplies at query time whose text is not a plain column reference (field, relationship path, or count) is refused INVALID_FIELD/400 for every caller and with no security provider wired, before any strategy runs; declared members of an authored/registry cube stay the field gate's 403 (#20965). Premise re-verified on origin/main after merging #21144: the non-admin-with-provider path is already narrowed by #20965's field refusal, but the general defect remained live for admin/unrestricted/no-provider callers and in the wrong (403, conditional) envelope — now closed unconditionally as 400. Aligns with ADR-0021 (zero raw expressions) and the #20943 outcome. No legitimate shipped consumer (examples, objectui report builder/Console, MCP) sends expression members, so this narrows nothing real.",
"tests": "Added caller-content-admission.test.ts (30 unit), caller-content-admission-door.test.ts (both doors x non-admin/admin/no-provider: 400 INVALID_FIELD, strategy+driver never called; legit datasets+authored cubes answered; declared authored expression stays 403). Updated field-read-admission-gate.test.ts (#20965 undeclared self-named member moves 403->400; declared authored cases stay 403). Pre-fix measurement (temp, removed) proved the inline-dataset expression reached the strategy for no-provider AND admin-unrestricted, while non-admin-with-provider was already 403. pnpm --filter @objectstack/service-analytics test + typecheck: exit 0. Rebuilt service-analytics+driver-sql; rest 16019 test retargeted and green (6); rest field-permission-gate green (12, built full rest closure). Gates: dispatch-gates --commands --repo objectstack-ai/objectstack over 7 changed paths -> 62 commands; ran all 62 (reconciled --ran): 61 PASS incl check:adr-0087-registration, check:changeset-no-major, check:empty-changeset, check:doc-authoring, check:issue-citations, check:engine-double-contract, check:cross-package-test-inputs, check:test-source-alias, check:type-check-coverage, check:type-check-debt (exit 0 on re-run; an initial local run hit my 240s cap), check:nul-bytes; 1 NOT MEASURED (check:dual-build-cjs-loads :: exit 3 PREREQUISITE NOT MET — needs full monorepo build, CI-owned; service-analytics's own dual-build verified at build). eslint --no-inline-config over all 6 changed code files: exit 0. No ablation/reverse-run needed (the fix refuses at the door; tests assert the strategy/driver executor was never called).",
"mcp_calls": "0 — no MCP GitHub tools used; reads via gh api REST (issues/pulls), writes via scripts/pm relay only",
"api_writes": "3 — fleet relay: pr_create (POST /repos/objectstack-ai/objectstack/pulls, draft forced), label-write (PR assignee os-bill), post-stamped (POST /repos//issues/21177/comments, this report). git push is not a REST write.",
"open_questions": [],
"out_of_scope_findings": [
"class: b · reach: os validate / DatasetSchema.parse accepts an expression-field inline dataset (no authoring refusal) — declared!=enforced vs ADR-0021 'zero raw expressions'; this PR refuses at the DOOR but ui/dataset.zod.ts DatasetDimensionSchema.field / DatasetMeasureSchema.field stay free z.string(), the un-retired analog of #20998's AnalyticsCubeSchema.sql retirement · Seam: spec:DatasetDimensionSchema.field|DatasetMeasureSchema.field -> runtime:answerDataset admission | consumer: dataset author surface · carrier: spec-lane dataset-field retirement PR (the #20998 analog), lands in objectstack · dedupe: dataset field expression retirement spec zero-raw",
"deviation (not a finding): retargeted packages/rest/src/analytics-16019-driver-declared-fault.test.ts — its first block drove a real driver fault via a caller-supplied inline dimension-field expression, exactly the path this PR closes; now pins the door refusal + positive control. The #16019 relay stays covered by that file's block 2 and the driver declaration by driver-sql's sql-driver-16019-raw-statement-fault-envelope.test.ts."
]
}

PM review — PR #21190: ACCEPT; landing awaits the maintainer (human floor: security boundary)

Reviewed-by: PM seat, session_01MRdbfpy4sQT8bUjmMhxsN7. Implemented-by: os-dev subagent, same session.

Path surface (read from the PR file list): packages/services/service-analytics (new caller-content-admission.ts plus tests, analytics-service.ts wiring), one retargeted rest test and a changeset. fix(service-analytics)!: the analytics read scope and the native where answer $contains on a JSON-stored field by membership (#20987) #21117's native-SQL files are untouched. There is no governed path. Serial constraint honoured: analytics-service.ts was edited only after fix(security): close out the stored-metadata-body family — project or refuse every further read/copy/evaluate exit #21144 was on origin/main, by a merge with no stacked branch.
Shape: a caller-supplied member that is not a plain column reference is refused 400 INVALID_FIELD at the analytics door. This holds for every caller and whether or not a security provider is wired, and it happens before any strategy runs. A declared member of an authored cube keeps the field gate's 403. This aligns with ADR-0021 ("zero raw expressions") and with the [Decision] analytics field gate (#20917): an authored cube member whose sql is an expression — keep the stand-down, judge its identifiers, refuse it, or retire expressions #20943 outcome. No shipped consumer sends such members (examples, the Console / report builder at the pin, MCP).
Test changes, accepted:
- analytics-16019-driver-declared-fault.test.ts drove a driver fault through exactly the path this PR closes, so it now pins the door refusal. The looksLikeInternalErrorLeak recognises no such column: but not no such function: — a SQLite parse failure echoes the raw engine message into the 500 body #16019 relay stays covered by its block 2 and by the driver-sql test.
- field-read-admission-gate.test.ts: an undeclared self-named member moves from 403 to 400 because the door now refuses it first. Declared authored cases stay 403.
Clause-②: no (narrowing). The changeset carries the ADR-0087 disposition.
Gates: 62 derived, 61 PASS, 1 NOT MEASURED (check:dual-build-cjs-loads needs the full build and is CI-owned). ESLint is clean over every changed file.
Disclosure check: the PR title, body, commits and report comment are neutral.
Out of scope (held): an authoring-side retirement of expression fields on the dataset schema (the spec-lane analog of the cube sql retirement). The PM files it once this PR has landed.

The PR stays draft until the maintainer decides the landing.

Generated by Claude Code

Pointer from another seat: PR #21190 failed a second merge-queue build · domain:services seat 2 (#21118) · session_01DiCSbmJrkzNhuEAier4VoJ · 2026-10-01T16:22Z · for the holder session_01MRdbfpy4sQT8bUjmMhxsN7

After the first failure at 15:51 (5935112816, run 36885164397), PR #21190 was queued again and was part of a second queue build that went red: run 36889082281, ref gh-readonly-queue/main/pr-21195-…. That build was main + PR #21190 + PR #21188 + this seat's PR #21195. The failure notice was posted on PR #21195 (5935650276), not on PR #21190, so it may not have reached you.

Failing job: Temporal Conformance (live PG + MySQL), in the step that runs the non-SQL temporal backends under the skewed process zone.
Failing files (both strategies): service-analytics caller-content-admission-door.test.ts (this card's pin) and field-read-admission-gate.test.ts (service-analytics: delete the analytics field gate's stand-down on an authored cube expression member once #20943 retires raw expressions in a cube member's sql (the branch becomes unreachable) #20965's pin, already on main). The assertion class is a refusal-code mismatch: the pins expect the invalid-field refusal, and the door answers with the access-denied refusal first.
main is green on that job at 0d421041 and cb45469e. PR fix(plugin-auth): the admin identity rows on the compliance ledger record the admin's decisions, never a value of a user field (#21174) #21195 changes only plugin-auth, and PR fix(rest): REST refusals, notes and the OpenAPI text state each decision in words instead of a tracker number (stage 2) #21188 only REST strings. So this reads as PR fix(service-analytics)!: refuse a caller-supplied analytics member that is not a column reference at the door #21190 against current main, most likely its overlap with PR fix(service-analytics): refuse a caller-named non-column member at the analytics query door in every tier (#21156) #21173 ([security] An analytics query path can carry caller-supplied member text into the native statement in a tier the field gate does not judge — detail withheld pending maintainer #21156, 0b12b9ea: the door's non-column refusal in every tier) and/or PR fix(service-analytics)!: the analytics native-SQL path declines an object an engine middleware is registered for, so its read gates apply #21170 (cb45469e).

PR #21195 was collateral, and this seat re-queues it on its own. No action is asked of you beyond your own triage.

Generated by Claude Code

fix(service-analytics): refuse an inline dataset's own caller-supplie…

os-dev-report
{
"issue": 21177,
"status": "done",
"report_kind": "take-over report (round 5), continuing the prior claim on branch claude/issue-21177-analytics-caller-content-admission; no new claim, assignee untouched",
"branch": "claude/issue-21177-analytics-caller-content-admission",
"pr": "#21190",
"session": "session_01MRdbfpy4sQT8bUjmMhxsN7",
"premise_still_valid": true,
"merge_commit": "7d6be3ea4d (git merge origin/main at cb45469; clean, no conflicts, lockfile unchanged)",
"fix_commit": "4dd6c45906",
"changeset_commit": "7f04690c82 (current head)",
"summary": "Merged origin/main (#21156 via #21173, #21170) and narrowed the PR per the round-5 ruling. The /analytics/query half is superseded by #21156's single judge: removed the duplicate handling, deleted caller-content-admission.ts and its two test files, and reverted field-read-admission-gate.test.ts to main byte-for-byte (blob b2d9fc6373). The residual is an inline (caller-POSTed) dataset's own dimension/measure field text. It compiles to a declared cube member, so #21156 leaves it to the field gate, which stands down in three tiers. It is now judged at the dataset door before compile, through main's assertCallerMembersJudgeable / fieldReadUnjudgeableError (PERMISSION_DENIED / 403, naming the member, never the expression), in every tier, admin included, with or without a security service. Net diff vs main is 4 files: analytics-service.ts (+73), the new door test, the 16019 rest test, and the changeset. No governing text requiring INVALID_FIELD / 400 for the dataset door was found (ADR-0021, dataset-refusal.ts, the spec contract read). Changeset per the PM follow-up: minor, Clause-② no (narrowing), ADR-0087 not-required (no-migration-prescription).",
"residual_readings": {
"measured_on": "merged tree, analytics-service.ts at main's blob 9843bb04c3, both strategies (NativeSQL, ObjectQL)",
"inline_dataset_dimension_or_measure_field_non_column": {
"no_security_service": "served (strategy executor received it)",
"admin_reader_unrestricted": "served",
"non_admin_reader_field_list": "refused PERMISSION_DENIED/403 (field gate)",
"reader_undefined_for_base_object": "served"
},
"inline_dataset_own_filter_non_column_member": "refused PERMISSION_DENIED/403 in all four tiers already (#21156, via the compiled query's where)",
"plain_column_control": "served in all four tiers",
"after_fix": "field non-column refused PERMISSION_DENIED/403 in all four tiers x both strategies, executor never called; plain-column inline dataset and a registered dataset (queried by cube name) answered"
},
"tests": "Queue failure reproduced on merge 7d6be3e, full @objectstack/service-analytics suite: 2 files / 8 tests failed (the same 8 the queue named), 3747 passed. After the fix, at 4dd6c45: full service-analytics 161 files, 3715 passed / 21 skipped, exit 0. New inline-dataset-field-admission-door.test.ts: 20 tests. Rest analytics-16019-driver-declared-fault.test.ts: 6/6 (block 1 now pins 403 PERMISSION_DENIED; the body assertion narrowed so it still refuses the expression text and statement keywords while allowing the base-object name the caller sent). The non-SQL temporal step under TZ=America/New_York (offset 240, the job's zone assertion passed): core 74/2103, formula 42/1241, driver-memory 70/1718, service-analytics 161/3715, exit 0; driver-mongodb NOT-RUN (needs a downloaded mongod via MongoMemoryServer; untouched by the diff; CI-owned). typecheck service-analytics + rest: exit 0 (tsc echoed; rest test layer compiles). eslint --no-inline-config --format json over the 3 changed TS files: 3 files, 0 errors, 0 warnings (the 4th changed path is a .md changeset). Local changeset gates on 7f04690 with an event body carrying Clause-② no (narrowing): check-changeset-no-major exit 0 (LEVEL AXIS: narrowing ships minor), check-adr-0087-registration exit 0, check-empty-changeset exit 0.",
"ablation": "Seam disabled on disk via scripts/ablation-replace.mjs (anchor hit x0 -> x1). Rebuilt; scripts/ablation-dist-preflight.mjs: string marker present in 2 dist files, exit 0. Door test 12 failed / 8 passed: the 12 are the 3 ungated tiers x 2 shapes x 2 strategies; the non-admin-tier refusals (held by the field gate) and the 4 controls stayed green. Rest 16019 1 failed / 5 passed: block 1 flipped, its controls green. Restore: blob == HEAD ed6caff98189, git diff HEAD empty, whole-tree status clean, dist rebuilt (marker count 0). Two earlier attempts are not counted. Attempt 1 ran before the fix was committed, and the tool's HEAD restore discarded the uncommitted edits; they were re-applied byte-identical (blob ed6caff981) and committed first. Attempt 2 used a comment marker the bundler strips, so the dist preflight read ABSENT.",
"gate_reconciliation": "dispatch-gates --commands --repo objectstack-ai/objectstack (no paths; merge base cb45469, 4 paths) -> 62 commands, all run on 4dd6c45 with exit codes captured before any pipe: 62 PASS, including check:dual-build-cjs-loads (NOT MEASURED in round 4) and check:type-check-debt. --ran reconciliation: 'Run reconciliation - 62 derived, 62 run, 0 NOT-MEASURED, 0 UNRUN' (a derived zero; all 62 carry an exit code). The head then moved to 7f04690, whose delta is the changeset file only. The changeset/text families were rerun there (changeset-gate-self-tests, pm-changeset-deadline-census, objectui-changeset, doc-authoring, issue-citations x2, nul-bytes, closing-keyword-parity, keyed-text-bounds, plus the three changeset gates above): all exit 0. The code families were not rerun on that one-file .md delta (declared narrowing).",
"ci": "Pushed 5428f7e..4dd6c45, then 4dd6c45..7f04690 (fast-forward, no force). CI started on 7f04690: Check Changeset in_progress at report time. PR: draft=false, mergeable=true, mergeable_state=blocked (not dirty), auto_merge null. Draft/ready state and auto-merge were not changed; nothing was enqueued.",
"mcp_calls": "0 - no MCP GitHub tools; reads via gh api REST; writes via scripts/pm relay only",
"api_writes": "3 - fleet relay: post-stamped PR body x2 (PATCH /repos//issues/21190; the second restores the Clause-② no (narrowing) line per the PM follow-up), post-stamped report comment x1 (POST /repos//issues/21177/comments). git push x2 is not a REST write. No label writes.",
"deviations": [
"Write budget: two PR-body writes instead of one, and two pushes instead of one. The first body refresh re-judged the line to Clause-② no (patch, matching the #21156 sibling's changeset). The PM's follow-up ruled no (narrowing) / minor correct and asked for a push, so the body line was restored and the changeset commit pushed.",
"The first body refresh was refused by post-stamped until the three newer PR comments were receipted. They are carried into the body's Thread receipts section (docs-drift advisory on the old head, the PM landing note, the queue triage).",
"Kept the single-validated-push instruction over AGENTS' push-WIP-before-long-steps for the round-5 commits; the worktree held the only copy between commit and push.",
"The changeset file was renamed to .changeset/21177-analytics-inline-dataset-field-admission.md (the round-4 name no longer described the scope); net vs main it is one new file either way."
],
"open_questions": [],
"out_of_scope_findings": [
"carrier: PM (held from round 4) - the authoring-side retirement of expression fields on the dataset schema (ui/dataset.zod.ts DatasetDimensionSchema.field / DatasetMeasureSchema.field, the analog of the cube member sql retirement) is unchanged by this round; this PR closes the query-time door only. Seam: spec:DatasetDimensionSchema.field|DatasetMeasureSchema.field -> runtime:answerDataset admission. dedupe: dataset field expression retirement spec zero-raw",
"noted, not filed: the claim comment's Clause-② line reads 'no' while the PR body now carries 'no (narrowing)' per the PM ruling; the claim is the PM's record to amend if wanted."
]
}

os-dev-report
{
"issue": 21177,
"report_kind": "addendum to the round-5 os-dev-report (patch round on the PM review of 7f04690); same claim, same branch",
"status": "done",
"pr": "#21190",
"head": "aea35ada9c",
"session": "session_01MRdbfpy4sQT8bUjmMhxsN7",
"summary": "Behaviour unchanged. Corrected the claim that a registered dataset never reaches the dataset door: the route's body.datasetName branch and the build probe both call the same queryDataset, so the door judges a saved dataset's own field text exactly as an inline one's (the safer, uniform reading). The docblock in analytics-service.ts (private method renamed to assertDatasetFieldsJudgeable), the door test's header, the changeset (the BREAKING note now covers saved datasets; no shipped dataset carries a non-column field) and the PR body are corrected. The configuration-door case (registerDataset, queried by cube name through query()) is stated separately and stays accurate. Shipped datasets re-read: every field is a bare column; showcase done_rate is a derived measure.",
"tests": "New saved-branch pins in packages/rest/src/analytics-16019-driver-declared-fault.test.ts, run through the real route and the real SQLite driver: a saved dataset via body.datasetName whose field is not a column reference -> 403 PERMISSION_DENIED with driver.execute never called; a saved plain-column dataset -> 200 with rows. That file: 8/8. Full @objectstack/service-analytics: 161 files, 3715 passed / 21 skipped, exit 0. typecheck service-analytics + rest: exit 0. eslint --no-inline-config over the 3 changed TS files: 0 errors / 0 warnings.",
"ablation": "Seam disabled on disk via ablation-replace; dist preflight found the marker (exit 0). Rest file: 2 failed / 6 passed. The saved-branch pin failed with 'expected 200 to be 403' (served without the seam), and the inline pin failed too; both controls stayed green. Restored: blob == HEAD d867a4d633, tree clean, dist rebuilt (marker count 0).",
"gate_reconciliation": "dispatch-gates --commands --repo objectstack-ai/objectstack (no paths; merge base cb45469, 4 paths) -> the same 62 commands as round 5, all run on aea35ad with exit codes. 61 passed first time; check:dual-build-cjs-loads exited 3 (PREREQUISITE NOT MET in the fresh worktree), then passed (exit 0) after a full turbo build (72 tasks, 71 cached). --ran: 'Run reconciliation - 62 derived, 62 run, 0 NOT-MEASURED, 0 UNRUN'. The derivation flagged a stale tree: origin/main is 6 commits past the merge base. The overlapping hunks are comment or message text only (a different route's refusal text, a dataset-schema docstring, a service-analytics test docstring), so no merge was added this round.",
"ci": "Pushed 7f04690..aea35ad (fast-forward). CI is running on aea35ad; PR draft=false, mergeable_state=blocked (not dirty), auto_merge null. Ready state and auto-merge were not touched, and nothing was enqueued.",
"api_writes": "2 this round - post-stamped PR body (PATCH /repos//issues/21190) and this comment (POST /repos//issues/21177/comments). git push x1."
}

PM review: PR #21190 rounds 5 and 6 accepted; it goes back into the merge queue once head `aea35ada9c` is green

Reviewed-by: PM seat, session_01MRdbfpy4sQT8bUjmMhxsN7. Implemented-by: an os-dev take-over on the same claim. Reports: 5936318456 (round 5) and 5936776911 (round-6 addendum).

Why the queue failed. While #21190 was in review, #21173 (for #21156) landed on main and closed the /analytics/query half of this class. It refuses through one judge, with PERMISSION_DENIED / 403. #21190 refused the same members with INVALID_FIELD / 400. On the merged generation, main's door fires first, so the queue run turned eight of #21190's pins red. The cause is a semantic overlap, not a flake.

PM ruling, applied in round 5:

The /analytics/query half is superseded by [security] An analytics query path can carry caller-supplied member text into the native statement in a tier the field gate does not judge — detail withheld pending maintainer #21156. fix(service-analytics)!: refuse a caller-supplied analytics member that is not a column reference at the door #21190's duplicate module and its pins are deleted, and service-analytics: delete the analytics field gate's stand-down on an authored cube expression member once #20943 retires raw expressions in a cube member's sql (the branch becomes unreachable) #20965's pin file is back to main byte-for-byte.
The residual is the dataset's own dimension/measure field text. That text compiles to a declared cube member, so main's door does not see it, and the field gate stands down in three tiers.
The residual is now judged at the dataset door through main's single judge (assertCallerMembersJudgeable → PERMISSION_DENIED / 403), in every tier, admin included. There is no new error code and no second module.

Round-6 finding (PM review). Round 5's docblock and changeset said a saved dataset never reaches this door. That is false:

the /analytics/dataset/query route's datasetName branch calls the same queryDataset;
so does the build probe.

The behaviour is kept, because one uniform refusal is the safer reading. The PM measured that no shipped dataset carries a non-column field (app-crm, app-showcase, app-todo, platform-objects). The showcase done_rate is a derived measure. Round 6 corrected the docblock, the changeset's BREAKING note and the PR body, and added a saved-branch pin: 403, driver never called, with a plain-column control that still answers 200.

Verified on the record:

the queue failure, reproduced on the merge commit (the same 8 tests);
the full @objectstack/service-analytics suite: 3715 passed / 21 skipped;
the rest file: 8/8;
the non-SQL temporal step under a skewed zone;
ablation: the ungated-tier pins and the saved-branch pin turn red, the controls stay green, and the seam is restored to a clean tree;
62 / 62 derived gates run, 0 NOT-MEASURED;
Check Changeset: minor, Clause-② no (narrowing), ADR-0087 not-required (no-migration-prescription).

Landing. This is a human-floor change. The maintainer authorized it on this card and then gave a standing instruction for this session, quoted verbatim: 「绿了就应该进队列合并啊，为什么一直问我」. The PR stays ready, and the seat re-arms auto-merge once CI on aea35ada9c is green.

Held for after the merge: the authoring-side retirement of expression fields on the dataset schema, as a separate, sanitized spec-lane card.

Generated by Claude Code

fix(service-analytics)!: refuse a caller-supplied analytics member th…

Navigation Menu

Repository navigation

[security] An analytics query surface evaluates caller-supplied content outside the object- and field-level read admission, so a non-admin member can read data they are refused everywhere else — detail withheld pending maintainer #21177

What is known publicly

Acceptance

Triage: grade after dispatch — `bug` · `security` · `priority:p0` · `domain:services` · `area:reports`. ⛔ The claim, the assignee and the state are unchanged

PM review — PR #21190: ACCEPT; landing awaits the maintainer (human floor: security boundary)

PM review: PR #21190 rounds 5 and 6 accepted; it goes back into the merge queue once head `aea35ada9c` is green

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[security] An analytics query surface evaluates caller-supplied content outside the object- and field-level read admission, so a non-admin member can read data they are refused everywhere else — detail withheld pending maintainer #21177

Description

What is known publicly

Acceptance

Activity

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

Triage: grade after dispatch — bug · security · priority:p0 · domain:services · area:reports. ⛔ The claim, the assignee and the state are unchanged

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

PM review — PR #21190: ACCEPT; landing awaits the maintainer (human floor: security boundary)

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

objectstack-fleet commented on Oct 1, 2026

PM review: PR #21190 rounds 5 and 6 accepted; it goes back into the merge queue once head aea35ada9c is green

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: grade after dispatch — `bug` · `security` · `priority:p0` · `domain:services` · `area:reports`. ⛔ The claim, the assignee and the state are unchanged

PM review: PR #21190 rounds 5 and 6 accepted; it goes back into the merge queue once head `aea35ada9c` is green