QA run · release-18.3.0-w4-l1 (40/40) · 7fa9873f · 2026-11-02 · 37 PASS / 2 PARTIAL / 1 FAIL / 0 BLOCKED / 0 NOT-RUN #29124

Activity

1. objectstack-fleet

   added

   qa-run

   on Nov 2, 2026
2. objectstack-fleet commented on Nov 2, 2026

   

   objectstack-fleetBot

   on Nov 2, 2026

   ContributorAuthor

   More actions

   domain:devx#1 · os-warren · session_01JQwxbZx1yZXiTrFXKbSdFZ

   Per-clause verdicts (1/2): access-security, ai, api-backend, approvals. Live legs only; the #28929 pass rows keep their verdict and are not repeated.

   access-security

   item · rev	clause	verdict	layer · text evidence
   access-security.fls-mask-and-strip · r8	A3	pass	browser-live (/_console, form-signed-in contributor C holding showcase_contributor): …/showcase_project/record/KRfrBH2I0Yqyh2MU/edit rendered (screenshot): input[name=budget] disabled=true value 90000, input[name=spent] disabled=true value 88000, each with "You do not have edit access to this field."; name/owner/start_date/end_date enabled. Admin contrast on the same form: budget and spent disabled=false. negative[0] held at the server: C's direct budget PATCH 403 "[Security] Field write denied … [budget]", re-read 90000
   access-security.fls-mask-and-strip · r8	A4	pass	HTTP-booted-server + browser-live: fresh member L holding only the STOCK showcase_client_liaison (+ baselines, no showcase_contributor): GET /api/v1/data/showcase_project/KRfrBH2I0Yqyh2MU 200 with budget, spent and budget_remaining keys ABSENT (not null); list 200, 0 of 6 rows carry any of the three keys; admin GET same row budget 90000, spent 88000, budget_remaining 2000. UI: L's detail page shows no Budget highlight and no Financials section, no 90,000/88,000/2,000 anywhere; its POST …/page network body carries the record with all three keys absent. Admin's detail shows Budget $90,000.00 and Spent $88,000.00 and its …/page record carries all three. UI and API agree for both personas (budget_remaining is not on the detail layout for either)
   access-security.grant-table-write-contract · r8	A4	pass	HTTP-booted-server (walled boot, OS_TENANCY_POSTURE=group, /auth/config tenancyPosture "group"): org A and org B created via /auth/organization/create, OA promoted to owner of A (sets organization_admin), NB/NB2 members of B only. OA in active org A: POST /api/v1/data/sys_user_permission_set {user_id: NB, permission_set: showcase_contributor} → 400 VALIDATION_FAILED fields[0] {field: user_id, code: reference_not_found, constraint.target sys_member}; NB2 same 400; controls: MA (member of A) 201, OA self 201; sys_user_permission_set rows for NB = 0
   access-security.me-permissions-aggregation-parity · r4	A0	pass	HTTP-booted-server: C map objects.showcase_project allowEdit:true → C PATCH name 200, admin re-read shows the new name; P map allowEdit:false → P identical PATCH 403 PERMISSION_DENIED, name unchanged. Fresh sign-in per persona; userId in each map = that persona's session user
   access-security.me-permissions-aggregation-parity · r4	A1	pass	HTTP-booted-server: C fields['showcase_project.budget'] {readable:true, editable:false} (spent and budget_remaining the same); C budget PATCH 999999999 → 403 "[Security] Field write denied … [budget]", admin re-read 90000 unchanged; C GET 200 carries budget 90000
   access-security.me-permissions-aggregation-parity · r4	A2	pass	HTTP-booted-server: registered apps showcase_app[req []], setup[req setup.access], account[req []]. /me/apps admin = [showcase_app, setup, account] (systemPermissions include setup.access); P = [showcase_app, account] (systemPermissions []). C, A and L match P. W = [showcase_app, setup, account] (holds setup.access). 0 returned apps violate requiredPermissions ⊆ systemPermissions or tabPermissions hidden, across admin/P/C/W/A/L
   access-security.me-permissions-aggregation-parity · r4	A3	pass	HTTP-booted-server (full stack, auth gate mounted), no Authorization header: GET /api/v1/auth/me/permissions 200 {"authenticated":false}; /api/v1/auth/me/localization 200 {"authenticated":false}; /api/v1/me/apps 200 {"apps":[]}. A garbage bearer also gets 200 {"authenticated":false} on permissions
   access-security.me-permissions-aggregation-parity · r4	A4	pass	HTTP-booted-server: P GET /auth/me/permissions?userId=‹C id› → userId = P's own id, body byte-identical to P's un-parameterised body, no showcase_contributor in permissionSets; ?userId= on /auth/me/localization and /me/apps also byte-identical to P's own answers
   access-security.me-permissions-aggregation-parity · r4	A5	pass	HTTP-booted-server: P /auth/me/localization 200 {authenticated:true, currency:null, locale:"en-US", timezone:"UTC"} (P's GET /api/settings/localization 403 SETTINGS_FORBIDDEN, so the settings surface is setup-gated and the resolved answer is not); admin PUT /api/settings/localization {currency:EUR, timezone:Asia/Shanghai} 200, P's fresh session then reads currency "EUR", timezone "Asia/Shanghai"; restored to currency null / timezone UTC afterwards
   access-security.me-permissions-aggregation-parity · r4	A6	pass	HTTP-booted-server: W (wall-less org admin via update-member-role admin; permissionSets [showcase_member_default, organization_admin_no_bypass, member_default], no admin_full_access): objects.showcase_semantic_zoo present allowEdit:true and a zoo read 200; objects.showcase_project allowEdit:true; W creates a project 201 and PATCHes its name 200; objects.sys_secret absent and GET /data/sys_secret 403 PERMISSION_DENIED. Only 4 of 115 registered objects are missing from W's map, all private: sys_secret, sys_flow_credential, sys_connector_credential, sys_connector_connection
   access-security.record-access-explain · r4	A0	pass	browser-live (/_console/studio/com.example.showcase/access, "Explain access", admin): user = A persona (showcase_auditor) / P persona, object showcase_private_note, record = probe 6PfNtH4HhTAzbzjh (owned by a third member M2). Screenshot shows the 10 layers (tenant_isolation, principal, required_permissions, object_crud, fls, owd_baseline, depth, sharing, vama_bypass, rls), each with its verdict badge. Record verdict for A: "Record … is VISIBLE … Decided by: View/Modify All bypass". For P: "… is NOT visible … Decided by: Sharing"
   access-security.record-access-explain · r4	A1	pass	HTTP-booted-server: admin POST /api/v1/security/explain {object: showcase_private_note, operation: read, userId: A, recordId: probe} 200, record {visible:true, decidedBy:"vama_bypass"}, layer vama_bypass verdict widens, contributors [{permission_set showcase_auditor, via "direct grant"}]
   access-security.record-access-explain · r4	A2	pass	HTTP-booted-server: A explain visible:true ∧ A's own GET /data/showcase_private_note/‹probe› 200. P explain visible:false ∧ P's own GET 404 RECORD_NOT_FOUND. The verdict agrees with the read for both personas
   access-security.record-access-explain · r4	A3	pass	HTTP-booted-server: plain-member explain 200, record {visible:false, decidedBy:"sharing"}, 10 layers, no EXPLAIN_FAILED. negative[0]: the admin asked, principal.userId = P with sets [showcase_member_default, member_default], vama_bypass not_applicable, so it is not visible because the admin asked
   access-security.record-access-explain · r4	A4	pass	browser-live + HTTP-booted-server: the panel's own POST /security/explain request (same body shape) answered a body byte-identical to the API-twin body for both personas. The DOM layer list (data-testid explain-layer-*) matches the body's layers in order and verdict for both (n/a ↔ not_applicable)
   access-security.share-link-landing-page · r6	A0	pass	browser-live + network, fresh anonymous context, /_console/s/‹plain token› (link_only, per-link redact [summary]): resolve 200, record keys id…title,project,status,owner_id, with internal_notes, deal_value and summary absent and none of their values in the body; redactedFields [internal_notes, deal_value, summary]. Page (screenshot) renders "Shared showcase_client_brief" with the fields, no redacted value in the DOM, and the notice "Some fields are hidden by the owner: internal_notes, deal_value, summary"
   access-security.share-link-landing-page · r6	A1	pass	browser-live + network: password link: 401 {code NEEDS_PASSWORD}, page "Password required"; wrong password via #share-pw + Continue → 401 {code WRONG_PASSWORD}, page re-prompts with "Wrong password."; correct → 200 and the record renders. Both 401 bodies carry no record and no secret value
   access-security.share-link-landing-page · r6	A2	pass	browser-live (screenshots) + network: revoked → resolve 410, page "Can't open this link / This link has expired or was revoked."; short-expiry link after lapse → 410, the same designed message; garbage token → 404, "This link is invalid or no longer available.". The same anonymous resolves over HTTP give 410 {EXPIRED_OR_REVOKED} / 404 {INVALID_OR_EXPIRED} with no record. None of the pages is blank, a spinner or raw JSON
   access-security.share-link-landing-page · r6	A3	pass	browser-live + network: signed_in link, anonymous fresh context: resolve 401 {code SIGN_IN_REQUIRED} with no record; page "Sign in required" + "The owner shared this link with signed-in users only."; the Sign in href is /_console/login?redirect=%2Fs%2F‹token›; 0 input[type=password]. Member P signed in through the console form (better-auth.session_token cookie + auth-session-token in localStorage): resolve 200 and the record renders
   access-security.share-link-landing-page · r6	A4	pass	browser-live (screenshot): a fresh context with no session loads /_console/s/‹plain token› standalone. It has 0 nav/aside elements and no sidebar or console chrome, and the URL stays /_console/s/‹token› with no login redirect

   ai

   item · rev	clause	verdict	layer · text evidence
   ai.mcp-coded-refusals-and-runtime-exposure · r4	A1	pass	HTTP-booted-server: admin osk_ tools/call get_record {objectName:showcase_task, recordId:does-not-exist-g2} → isError:true, text {error:{code:"RECORD_NOT_FOUND",message:"Record does-not-exist-g2 not found in showcase_task",status:404}}; the id spelling → -32602 naming id → recordId
   ai.mcp-coded-refusals-and-runtime-exposure · r4	A2	pass	HTTP-booted-server: PUT /meta/object qa_hidden_obj (enable.apiEnabled:false, sharingModel) and qa_shown_obj 200; query_records qa_hidden_obj → {error:{code:OBJECT_API_DISABLED,message:"object is not exposed via the API",status:404}}, qa_shown_obj → records [] total 0; boot stopped (group SIGTERM) and re-booted on the same file DB: the first call after the boot got the same OBJECT_API_DISABLED 404, and qa_shown_obj was still served; after DELETE /meta/object (200 both): list_objects no longer lists either, query → OBJECT_NOT_FOUND 404, describe → 'Object "qa_hidden_obj" not found'
   ai.mcp-coded-refusals-and-runtime-exposure · r4	A3	pass	HTTP-booted-server: staged qa_ai_mark (PUT /meta/action 200, declarative update progress:50, ai.exposed); auditor.demo osk_ run_action on task 1rBi5zPB9S_3Ik2h → isError, body keys ["error"] → ["code","message","status"] = PERMISSION_DENIED / generic message / 403; scanning the raw body found no position, permissionSet, details, grant or profile key; the task's progress stayed 100. Admin control on task 5HO3aK-Qql7q-HBd → ok:true, progress 45→50
   ai.mcp-http-surface · r8	A0	pass	HTTP-booted-server (OS_MCP_SERVER_ENABLED=false boot :41622): GET /mcp/skill as anon and as a session caller → 404 {code:RESOURCE_NOT_FOUND,message:"MCP server is not enabled"}; POST /mcp as a session bearer and as an osk_ caller → the same 404; anonymous POST /mcp → 401 UNAUTHENTICATED "Unauthorized: a valid API key is required", with no WWW-Authenticate header. Observed the 404 (switch off), never the 501. N1 held
   ai.mcp-http-surface · r8	A1	pass	HTTP-booted-server (default boot :41621): anon GET /api/v1/mcp/skill → 200, content-type "text/markdown; charset=utf-8", content-disposition 'inline; filename="SKILL.md"', cache-control no-store; the body starts with front-matter name: objectstack, and its "## Connect" block carries http://localhost:41621/api/v1/mcp
   ai.mcp-http-surface · r8	A3	pass	HTTP-booted-server: OAuth-live loopback boot: anon POST /mcp initialize → 401 UNAUTHENTICATED "Unauthorized: a valid OAuth access token or API key is required" + WWW-Authenticate 'Bearer realm="ObjectStack MCP", resource_metadata="http://localhost:41621/.well-known/oauth-protected-resource"'; the OAuth-dark boots (MCP-off :41622, public 192.0.2.2 :41624) → "Unauthorized: a valid API key is required", no header. N0: anon tools/call query_records and tools/list → 401 before any tool ran
   ai.mcp-http-surface · r8	A4	pass	HTTP-booted-server: admin osk_ (minted via POST /api/v1/keys 201): initialize 200 protocolVersion 2025-03-26, capabilities {prompts,tools}; tools/list = 12 names, exactly the documented set (list_objects, describe_object, validate_expression, query_records, aggregate_records, get_record, create_record, update_record, delete_record, list_actions, run_action, resume_run): none missing, none extra, no degradation
   ai.mcp-http-surface · r8	A5	pass	HTTP-booted-server: real embedded-AS token (native DCR + auth-code + PKCE + consent), scope "openid profile" → initialize 403 PERMISSION_DENIED "Forbidden: the access token grants none of the MCP scopes (data:read, data:write, actions:execute, metadata:write)" + WWW-Authenticate 'Bearer error="insufficient_scope", scope="data:read data:write actions:execute metadata:write", resource_metadata=…'; a data:read token → tools/list = list_objects, describe_object, validate_expression, query_records, aggregate_records, get_record only, and create_record → "Tool create_record not found"
   ai.mcp-oauth-private-host-transport · r6	A0	pass	HTTP-booted-server: boot (a) OS_AUTH_URL=http://10.0.0.5:41623. The address was NOT held: the item's fixtures.requires[0] (r6) says an unheld literal stages this clause in full. Both PRM paths → 200 {resource:"http://10.0.0.5:41623/api/v1/mcp", authorization_servers:["http://10.0.0.5:41623/api/v1/auth"], 4 MCP scopes+offline_access}; anon /mcp 401 resource_metadata="http://10.0.0.5:41623/.well-known/oauth-protected-resource". The advertised origin is the private one, never localhost (negative[1])
   ai.mcp-oauth-private-host-transport · r6	A1	pass	HTTP-booted-server: boot (b) OS_AUTH_URL=http://192.0.2.2:41624 (TEST-NET-1, PUBLIC to the rule; held by this container, so every request went to 192.0.2.2 itself): both PRM paths → 404 ENDPOINT_NOT_FOUND; anon /mcp 401 with no WWW-Authenticate; the boot log has 1× "OAuth discovery is served over PUBLIC plain HTTP" and 1× "MCP server is enabled but the OAuth track is NOT live (base URL fails the OAuth 2.1 transport rule …)", and 0× "OAuth is served UNENCRYPTED". N0 and N2 held
   ai.mcp-oauth-private-host-transport · r6	A2	pass	log (captured boot logs): "OAuth is served UNENCRYPTED": (a) 1 at WARN; (c) 1 at INFO in the --log-level info boot and 0 in the default warn-level re-boot on the same port; (b) 0; https 0. "OAuth discovery is served over PUBLIC plain HTTP": (b) 1 (the line states the AS discovery surface, MCP OAuth track DISABLED, TLS the remedy); (a) 0, (c) 0, https 0. Issuers named: http://10.0.0.5:41623/api/v1/auth, http://localhost:41621/api/v1/auth, http://192.0.2.2:41624/api/v1/auth; 「OAuth 未加密」 0 everywhere
   ai.mcp-oauth-private-host-transport · r6	A3	pass	HTTP-booted-server: boot (c) loopback :41621, both PRM paths → 200 {resource:"http://localhost:41621/api/v1/mcp", authorization_servers:["http://localhost:41621/api/v1/auth"], same scopes_supported/bearer_methods/resource_name}, the same shape as (a) with the loopback origin in place of the private one; anon 401 carries the loopback resource_metadata pointer
   ai.mcp-oauth-private-host-transport · r6	A4	pass	HTTP-booted-server: boot (b) re-run as :41625, OS_AUTH_URL=http://192.0.2.2:41625 plus OS_ALLOW_INSECURE_OAUTH_HTTP=true (confirmed in the listener's /proc environ): both PRM paths 404 again, anon 401 with no pointer, the same PUBLIC + NOT-live lines, UNENCRYPTED 0; git grep OS_ALLOW_INSECURE_OAUTH_HTTP finds it only in ai.json, plugin-auth CHANGELOG.md and two test files
   ai.mcp-validate-expression · r6	A0	pass	HTTP-booted-server: validate_expression showcase_task 'record.estimate_hours * 1.1' formula → ok:true, inferredType "number", errors [] warnings [], inScope {dialect,roots,fields(26, incl. estimate_hours),functions(38)}
   ai.mcp-validate-expression · r6	A1	pass	HTTP-booted-server: 'estimate_hours * 1.1' formula → ok:false, error code bare-reference "… Write record.estimate_hours."; 'record.estimate_hourz * 1' → ok:false, error unknown-field "did you mean estimate_hours?" (params.suggestion); 'record.title * 2' → ok:true, warning type-mismatch string * int, errors []
   ai.mcp-validate-expression · r6	A2	pass	HTTP-booted-server: the same 'estimate_hours > 8' → flow_condition ok:true errors [] (no inferredType); formula ok:false bare-reference error (inferredType boolean)
   ai.mcp-validate-expression · r6	A3	pass	HTTP-booted-server: objectName no_such_obj → isError 'Object "no_such_obj" not found'; sys_user → isError 'Object "sys_user" is a system object and is not exposed via MCP' (text only, no schema); N1: '' and ' ' at formula, and '' at template, → ok:false with one invalid_union error each (the EVALUATED_EXPRESSION_SOURCE_REQUIRED sentence, the template sentence), no inferredType
   ai.skill-instructions-mcp-prompts · r4	A0	pass	HTTP-booted-server: PUT /meta/skill qa_prompt_full / qa_prompt_bare / qa_prompt_off (active:false) → 200 each; osk_ initialize capabilities include prompts; prompts/list = [{name:qa_prompt_full,title:"QA Prompt Full",description:"Full QA skill"}]
   ai.skill-instructions-mcp-prompts · r4	A1	pass	HTTP-booted-server: prompts/get qa_prompt_full → messages[0].content.text = "QA-INSTR-FULL-V1: always answer in haiku." (the authored text); after re-PUT with V2 instructions the next get answered "QA-INSTR-FULL-V2: answer in limericks." (read again at get time, no restart)
   ai.skill-instructions-mcp-prompts · r4	A2	pass	HTTP-booted-server: before: prompts/list omits qa_prompt_bare and qa_prompt_off (N0 held); re-PUT qa_prompt_off active:true → GET /meta/skill/qa_prompt_off reads active:true → prompts/list now [qa_prompt_full, qa_prompt_off], prompts/get serves "QA-INSTR-OFF"
   ai.skill-instructions-mcp-prompts · r4	A3	pass	HTTP-booted-server: prompts/get qa_prompt_bare (no instructions), qa_prompt_off while inactive (N1 held) and no_such_skill → JSON-RPC error -32602 'Prompt "…" not found' each, never an empty success

   api-backend

   item · rev	clause	verdict	layer · text evidence
   api-backend.action-activation-door-contract · r7	A4	blocked(fixture, knownGaps[3])	HTTP-booted-server: the 404 and no-row halves held live: admin POST /actions/_activation/showcase_task/no_such_action {"enabled":false} → 404 RESOURCE_NOT_FOUND "…has no declaration — there is nothing to switch off…(ADR-0126 §4)" (same on no_such_object and global); member Mei's mis-bodied undeclared probe → 403 PERMISSION_DENIED (manage_metadata sentence); sys_metadata_activation 0 rows before and after every refusal (node:sqlite count 0; GET /data/sys_metadata_activation total 0). The metadata-plane 503 arm (declaration.degraded, only when resolveRouteActionDeclaration throws, action-execution.ts:4490) has no stock door and no pin
   api-backend.bulk-write-contract · r6	A0	pass	HTTP-booted-server: member m1 POST /data/showcase_private_note/createMany bare array of 3 → 201 {object, records[3]}, each owner_id/created_by/updated_by = m1, organization_id = Default Org, pinned default false; {records:[…]} wrapper → 400 VALIDATION_FAILED fields[0] {records, invalid_type}. Single-door POST row: GET key sets identical (12 keys) and stored non-null column sets identical (body,created_at,created_by,id,organization_id,owner_id,pinned,title,updated_at,updated_by)
   api-backend.bulk-write-contract · r6	A1	pass	HTTP-booted-server: m1 updateMany {records:[{id,data:{body}}×3 markers]} → 200 total 3 succeeded 3; admin re-read: 3/3 markers body 'upd-r1', control single note body 'control' and m2's note body 'foreign-orig' unchanged; {where,data} → 400 VALIDATION_FAILED fields[0] {records, required}
   api-backend.bulk-write-contract · r6	A2	pass	HTTP-booted-server: m1 updateMany nonexistent id → 200 success:false succeeded 0, row RECORD_NOT_FOUND; m2's foreign note id → 200 success:false, row RECORD_NOT_FOUND (same code, same message shape); both in one call under continueOnError:true → both RECORD_NOT_FOUND; foreign row body/updated_at identical before/after (node:sqlite); records:[] → 200 total 0; per-object /batch update own+foreign (continueOnError) → own success, foreign RECORD_NOT_FOUND; admin repeat deleteMany of deleted ids → row0 RECORD_NOT_FOUND, rest NOT_ATTEMPTED. N0 held (foreign row neither counted nor mutated)
   api-backend.bulk-write-contract · r6	A3	pass	HTTP-booted-server: admin createMany 201 records → 400 BATCH_TOO_LARGE {count:201,max:200}; per-object /batch create 201 → 400 BATCH_TOO_LARGE; that batch's marker rows in DB = 0; control createMany 200 → 201
   api-backend.bulk-write-contract · r6	A4	pass	HTTP-booted-server: m1 deleteMany 2 own ids → 200 row0 PERMISSION_DENIED, row1 NOT_ATTEMPTED, 3/3 markers still present; m1 deleteMany foreign id → PERMISSION_DENIED (object gate first, as steps[5] says); admin deleteMany 3 marker ids → 200 succeeded 3; GET each → 404; single control note and foreign note → 200 (remain); {where} body → 400 fields[0] {ids, required}
   api-backend.bulk-write-contract · r6	A5	pass	HTTP-booted-server: grep ERR_BULK_RESULT_MISMATCH over every captured response and boot log → 0 hits (positive control: BATCH_TOO_LARGE 2 hits in the same capture)

   approvals

   item · rev	clause	verdict	layer · text evidence
   approvals.act-door-and-admin-rescue-slate · r5	A1	pass	HTTP-booted-server (os dev self-hosted mount) + pins: PUT /api/v1/approvals/act?token=garbage → 404 application/json {error.code ENDPOINT_NOT_FOUND}, no Allow header (byte-identical to the unmounted control PUT /approvals/no_such_door_xyz); GET and form POST token=garbage → 200 text/html "Invalid link · 链接无效", no summary rows; DELETE/PATCH also 404 ENDPOINT_NOT_FOUND; requests unchanged before/after. Live-token GET-renders/POST-redeems halves pin-scored per verify: hono-approvals-act.conformance 4 + action-page-member 12 passed at this subject (#28929 A0)
   approvals.email-action-token-door · r10	A0	blocked(fixture, knownGaps[0])	HTTP-booted-server (live at-rest leg held) + pins: admin POST /approvals/requests/‹invoice req›/remind → 200, actions gained one remind row (actor Dev Admin); the approval.reminder sys_notification row's payload = {title, body, actionUrl:"/system/approvals?request=…"} — no actions key, no /approvals/act URL, no token=; sys_approval_token 0 rows (columns token_hash…, no raw column); GET /data/sys_approval_token → 404 OBJECT_API_DISABLED (designed). Pins passed (#28929). The non-pending-approver FORBIDDEN arm has no pin and no door: issueActionTokens has no production caller at the subject (re-grepped)
   approvals.email-action-token-door · r10	A1	blocked(fixture, knownGaps[0])	none live: the byte-identity of the actions timeline and the token's consumed_at across repeated GETs needs a minted raw token; no stock door mints one (issueActionTokens' only references are its definition; /approvals/act producers are routing only). Pins for the status/decision half passed (#28929)
   approvals.email-action-token-door · r10	A3	blocked(fixture, knownGaps[0])	none live: the consume-FIRST half (a decide that fails still burns the token) needs a raw token; replay halves pin-held (#28929)
   approvals.email-action-token-door · r10	A4	blocked(fixture, knownGaps[0])	HTTP-booted-server (invalid shape held): clean anonymous client (GET /approvals/requests → 401) GET and form POST /approvals/act?token=garbage → 200 text/html "Invalid link · 链接无效", body has only the badge, "This link is not recognized" and the inbox link — no summary rows; 3 pending requests + their actions byte-identical before/after. The not_pending / not_approver PAGES with summary rows need a live token (no door, no page-level pin)
   approvals.email-action-token-door · r10	A5	blocked(fixture, knownGaps[0])	none: no raw token exists to present as Bearer; converse pin (token is the only credential) passed (#28929)
   approvals.request-page-reply-and-reader-visibility · r5	A1	pass	HTTP-booted-server: GET /meta/page/sys_approval_request_detail → 200 with 1 'record:approval_decision'; on seeded expense request areq_97a8…: {body} → 400 VALIDATION_FAILED 'comment is required'; Ada Auditor (pending approver) and Mei Phone (submitter) POST …/comment {comment} → 200 each; GET …/actions comment rows 0 → 2 (actor auditor, actor phone); member m2 (neither) → 403 FORBIDDEN "actor … is not on this request", no row
   approvals.request-page-reply-and-reader-visibility · r5	A2	pass	HTTP-booted-server on a run-authored grant (scratch package com.qa.g3reader: permission set qarv_req_reader {sys_approval_request: allowRead} via position qarv_reqreader bound to member m3): m3 GET /data/sys_approval_request/‹expense req› → 404 RECORD_NOT_FOUND, body identical (modulo id) to a nonexistent id's 404; m3 list → 200 records [] total 0 ($count 0) vs admin total 12. No-grant leg: m2 and Mei (submitter-participant) by-id and list → 403 PERMISSION_DENIED
   approvals.request-page-reply-and-reader-visibility · r5	A3	pass	HTTP-booted-server on a run-authored object (qarv_doc in scratch package com.qa.g3reader, sharingModel public_read, enable.approvalsVisibleToReaders) + scratch flow w4r_reader; reader m2 (read via position qarv_reader): flag ON → GET /approvals/requests?object=qarv_doc&recordId=DOC lists the request, by-id 200 viewer {can_act:false,is_submitter:false,can_override:false}, actions [submit]; inbox (no record) [] and my-pending []; approve → 403 FORBIDDEN, comment → 403; m1 (no record read) list [] by-id 404. PUT object flag OFF (no restart) → m2 record list [], by-id 404, actions 404; flag ON again → visible again
   approvals.request-page-reply-and-reader-visibility · r5	A4	pass	browser-live (Playwright chromium, form sign-in, console from packages/console/dist): /_console/apps/setup/sys_approval_request/record/areq_97a8… as Ada Auditor (pending approver): decision panel renders — "Sign-off progress — 0 of 2 groups" (finance 0/1, manager 0/1), "Waiting on" Dev Admin · manager / Ada Auditor (demo) · finance, buttons Approve / Reject / Reassign / Send back / Request info / Reply, Timeline 3
   approvals.status-mirror-field · r6	A0	pass	HTTP-booted-server: scratch flow w4m_r1 (POST /automation; approval node approvers [{type:user, admin}], approvalStatusField f_textarea, lockRecord true, maxRevisions 1) on showcase_field_zoo, submitter m1 (contributor-bound): pending (updated_by m1) → approve 'approved' (updated_by admin); reject 'rejected'; submitter recall 'recalled' (updated_by m1); revise 'returned' (admin); resubmit → 'pending' (m1); second revise past maxRevisions → request rejected, field 'rejected'
   approvals.status-mirror-field · r6	A1	pass	HTTP-booted-server: record X_GC5-QPm7noYtjG pending with f_textarea='pending' already landed; submitter m1 PATCH {f_email} → 409 RECORD_LOCKED "…locked while an approval is in progress…"; f_email stayed null, updated_at unchanged
   approvals.status-mirror-field · r6	A2	pass	HTTP-booted-server (wall-clock offset shim per approvals.sla-escalation fixtures) + pin: boot 41632 with NODE_OPTIONS --require shim; flow w4s_sla (escalation auto_approve, timeoutHours 1, mirror f_textarea) + record_change cascade w4c_cascade (runAs default 'user', stamps f_email). Human approve → mirror 'approved', cascade run completed user_id = decider, f_email 'cascaded@example.com'. Offset +2h02m, restart → boot sweep: request approved, actions [submit, escalate actor null 'auto_approve', approve actor null 'SLA escalation'], mirror 'approved', cascade run FAILED user_id null "[runAs] refusing a data operation … no trigger user could be resolved", f_email null. Dead-run half: status-mirror-cascade 2 passed (#28929)
   approvals.status-mirror-field · r6	A3	pass	HTTP-booted-server: flow w4n_r1 (no approvalStatusField): record created f_textarea='seed-value' stayed 'seed-value' and updated_by m1/updated_at unchanged through submit+approve; second record f_textarea null stayed null through submit+reject
   approvals.status-mirror-field · r6	A4	pass	HTTP-booted-server + log: flow w4b_r1 (approvalStatusField f_does_not_exist): approve → 200 {finalized:true, decision approve, resumed:true}; request approved; actions [submit m1, approve admin]; run completed; boot log "WARN [approvals] mirrorStatusField failed: Unknown field 'f_does_not_exist' on object 'showcase_field_zoo'" (twice: open and decide)

   ---

   Generated by Claude Code
3. objectstack-fleet commented on Nov 2, 2026

   

   objectstack-fleetBot

   on Nov 2, 2026

   ContributorAuthor

   More actions

   domain:devx#1 · os-warren · session_01JQwxbZx1yZXiTrFXKbSdFZ

   Per-clause verdicts (2/2): attachments-storage, automation, cli. Live legs only; the #28929 pass rows keep their verdict and are not repeated.

   attachments-storage

   item · rev	clause	verdict	layer · text evidence
   attachments-storage.attach-requires-parent-edit · r9	A2	pass	HTTP-booted-server: C-uploaded attachment on qa_shared: A (reads, no edit) DELETE 403 ATTACHMENT_DELETE_DENIED, B (no read) DELETE 404 RECORD_NOT_FOUND, row survives both (admin GET 200, DB count 1); uploader C with parent edit removed (row re-owned to admin; C PATCH row 403, C read 200) DELETE own attachment 200, row gone. Persona C rebound qa_att_editor→qa_att_reader first (see Accuracy: editor set has no sys_attachment delete, so RBAC 403 PERMISSION_DENIED answered ahead of the hook)
   attachments-storage.attach-requires-parent-edit · r9	A5	pass	HTTP-booted-server: admin PATCH sys_attachment {parent_object:qa_nofiles,parent_id:N} 403 FILES_DISABLED; re-read (API + DB) parent_object qa_shared / parent_id unchanged; half-applies refused too (parent_object only 403 FILES_DISABLED, parent_id only 403 ATTACHMENT_PARENT_ACCESS, row unchanged); control re-point onto files-enabled qa_vault 200 (N2 held)
   attachments-storage.download-authz-both-sides · r10	A2	pass	HTTP-booted-server: admin file on admin-owned qa_shared: A and C /url 200 {success:true,data:{url}}, url followed 200 byte-equal; B 403 ATTACHMENT_DOWNLOAD_DENIED. Uploader bypass: C's own file attached to C-created qa_vault row then re-owned to admin (C GET row 404): C /url 200, bytes equal, /files 302; A/B 403 ATTACHMENT_DOWNLOAD_DENIED; gated grant TTL 300 s vs presign 3600 s (exp read from the minted token)
   attachments-storage.download-authz-both-sides · r10	A4	pass	browser-live (built /_console, 1600px, three-column, panel resident): persona A form sign-in (get-session = qa-vault-a), qa_shared record panel listed 4 files; admin DELETE A's qa_att_reader binding 200 without reload; Download click → GET /api/v1/storage/files/ID/url 403 ATTACHMENT_DOWNLOAD_DENIED; panel role=alert "You don't have access to download this attachment." (no "Download failed", no JSON; N1 held); A re-bound 201
   attachments-storage.inline-grid-receipt-cells · r5	A2	pass	browser-live (built /_console, twice on fresh loads): picking receipt.png fired POST /api/v1/storage/upload/presigned 200, PUT /api/v1/storage/_local/raw/TOKEN 200, POST /api/v1/storage/upload/complete 200 (then GET /files/ID 302 → raw 200 for the thumbnail); chip file-cell-chip "receipt.png" with a remove button before submit
   attachments-storage.inline-grid-receipt-cells · r5	A3	pass	browser-live + file-DB read (twice): POST /api/v1/batch 200, body {results:[invoice,line]} no errors; line op receipt = bare uuid string (not data:/blob:/object, not the name; N1 held); sys_file row status committed, ref_object showcase_invoice_line, ref_id = the created line id, ref_field receipt, acl private
   attachments-storage.orphan-tombstone-reap · r4	A1	pass	HTTP-booted-server scheduler sweep (restart, initial sweep at +60 s): tombstone T1 (deleted_at -31d) row gone AND bytes absent; pending P1 (created_at -8d) row + bytes gone. Failed byte delete staged by replacing T2's byte file with a directory: served log "WARN [storage] reap guard: reclaim failed for sys_file ID (EISDIR: illegal operation on a directory, unlink …); retrying next sweep", T2 row kept status deleted. Blocker removed + restart: next sweep reaped T2 row and bytes (retry proven). Fresh tombstone T3 and committed control unchanged
   attachments-storage.presigned-upload-roundtrip · r3	A0	pass	HTTP-booted-server: presign 200 {success,data:{uploadUrl,method:PUT,headers,fileId,expiresIn:3600,downloadUrl}}; GET /data/sys_file/ID status pending, owner_id = admin id despite body owner_id/ownerId SPOOFED_OWNER (DB agrees, scope attachments, acl private)
   attachments-storage.presigned-upload-roundtrip · r3	A1	pass	HTTP-booted-server: sys_file status pending before POST /upload/complete (200), committed after (API and DB reads)
   attachments-storage.presigned-upload-roundtrip · r3	A2	pass	HTTP-booted-server: /files/ID/url 200 {success:true,data:{url}} (keys exactly success,data / url); GET url 200, 35 bytes, sha256 equal to upload; presign downloadUrl followed 200 (N1 held)
   attachments-storage.presigned-upload-roundtrip · r3	A3	pass	network (HTTP-booted-server): GET /files/ID 302, Location identical to the /url signed url; following it 200 byte-equal
   attachments-storage.presigned-upload-roundtrip · r3	A4	pass	HTTP-booted-server: anonymous presign 401 {success:false,error:{code:AUTH_REQUIRED}}; sys_file count 1→1; no "[storage] no session resolver" line in the boot log (N0 held)
   attachments-storage.read-inherits-parent-rls · r9	A3	pass	HTTP-booted-server + served log: ordinary half: 2309 sys_attachment rows (visible rows at both ends, 2300 vault filler between) — A and C served 6/6 incl. first+last, total = rows served, vault rows 0; admin 2309/2309; no '-row cap' / 'attachment read visibility' line in the log. Probe half reached live via a runtime federated parent (qa_ext on showcase_external, enable.files) with 2105 rows: log "WARN [storage] attachment read visibility: the bounded probe's candidate pre-scan (federated-parent: qa_ext, …) hit the 2000-row cap; the visibility filter for these rows is fail-closed and may omit visible rows"; A served 2100 (2 visible rows outside the window omitted, 3 dangling excluded, vault 0) — exclusion, no leak; B (no qa_ext read) 0 qa_ext rows
   attachments-storage.sys-file-status-pipeline · r5	A0	pass	HTTP-booted-server reads (GET /data/sys_file): pending after presign → committed after complete → committed after DP join delete → deleted (deleted_at set) after last join delete → committed (deleted_at null) after re-attach
   attachments-storage.sys-file-status-pipeline · r5	A1	pass	HTTP-booted-server: pending file /url 404 FILE_NOT_FOUND and /files 404 FILE_NOT_FOUND (no Location); also after the PUT, before complete: /url 404 FILE_NOT_FOUND
   attachments-storage.sys-file-status-pipeline · r5	A2	pass	HTTP-booted-server: holder-less tombstone /url and /files 404 FILE_NOT_FOUND; deleted row + live join row inserted behind the hooks (driver write): /url 200 bytes equal, /files 302, sys_file status/deleted_at/updated_at byte-identical before and after (tombstone not rewritten); holder removed → 404 again; deleted + ref_* owner stamp (driver) → /url 200; hooked re-attach → row committed/deleted_at null BEFORE any download, then /url 200 bytes equal
   attachments-storage.sys-file-status-pipeline · r5	A5	pass	HTTP-booted-server: user-scope file joined to Website Relaunch then join deleted → committed, deleted_at null before and after; user-scope field-owned file (qa_shared.doc, ref_object stamped) joined then join deleted → unchanged
   attachments-storage.upload-limits-scope-and-ownership · r4	A1	pass	HTTP-booted-server: 101 MiB presign 413 PAYLOAD_TOO_LARGE naming 100 MB, sys_file 7→7 (99 MiB control 200); PUT /api/settings/storage max_upload_mb=1 200; 2 MiB presign 413 naming 1 MB without restart, count 8→8; chunked init 2 MiB 413; 0.5 MiB 200; restored to 100
   attachments-storage.upload-limits-scope-and-ownership · r4	A2	pass	HTTP-booted-server: scope avatars / PUBLIC / public / null → 400 INVALID_REQUEST naming user, tenant, private, temp, attachments ('public' gets the "never made a file publicly readable" sentence); omitted → 200, sys_file scope user
   attachments-storage.upload-limits-scope-and-ownership · r4	A5	pass	HTTP-booted-server + driver write: qa_shared gained file fields doc/doc2; raw UPDATE planted R1's file id into R2.doc2 (raw column read back holds it); reader D (qa_shared read only; GET sys_file 403) GET R2: doc {id,name,size,mimeType,url}, doc2 {id, metadataRefused:true}; admin (sys_file read) gets full metadata for doc2 as the step predicts
   attachments-storage.upload-session-abort · r8	A0	pass	HTTP-booted-server + file-DB reads: init 200 {uploadId,resumeToken,fileId,totalChunks 2,chunkSize 5242880,expiresAt}; row in_progress 0 → PUT 0 (eTag) row 1, progress {uploadedChunks 1, percentComplete 83, status in_progress} → PUT 1 row 2 → complete 200, row completed 2/2, sys_file committed; progress then status completed
   attachments-storage.upload-session-abort · r8	A3	pass	HTTP-booted-server scheduler sweep + served log: abandoned session (expires_at -2d) with its parts dir made immutable (chattr +i): log "WARN [storage] reap guard: multipart abort failed for sys_upload_session ID (ENOTDIR …); retrying next sweep", row kept in_progress with backend_upload_id; sibling abandoned session reaped and parts gone; attribute removed + restart → next sweep reaped the row and its parts (N0 held)
   attachments-storage.upload-session-abort · r8	A4	pass	HTTP-booted-server: settled statuses read in_progress, completed, failed, expired. 'failed' driven live (part file removed from the backend before complete → 500 INTERNAL, row failed; re-PUT + re-complete → completed); bogus eTag 409 RESOURCE_CONFLICT, row stays in_progress. Retention: completed/failed/expired rows with created_at -8d and expires_at not past TTL all reaped; in_progress control with created_at -8d kept (onlyWhen holds). 'completing' transient per knownGaps[1]
   attachments-storage.upload-session-abort · r8	A5	pass	HTTP-booted-server: expires_at backdated 60 s (driver, no sweep): chunk PUT 410 UPLOAD_SESSION_EXPIRED, row expired, uploaded_chunks 1 unchanged; complete 410 UPLOAD_SESSION_EXPIRED; progress 200 status expired; in-deadline session chunk PUT 200

   automation

   item · rev	clause	verdict	layer · text evidence
   automation.trigger-type-matrix · r10	A0	pass	HTTP-booted-server (+browser for screen): one run per variant, trigger = launch event: record_change run_fc104edc showcase_urgent_task_alert (Flow.type record_change) {type:record-after-write, object:showcase_task, recordId:kyy-DTdhm-p0szAD}; autolaunched run_e12f4c42 showcase_batch_reminders (autolaunched) {type:manual}; schedule run_153ee789 showcase_scheduled_digest (schedule) {type:schedule}; screen run_706616fa showcase_reassign_wizard (screen) {type:manual, object:showcase_task, recordId:1uVksVze6EflHY2B}; api run_fe097e72 showcase_inbound_task_webhook (api) {type:api}; time_relative run_2f8d2a56 qa_g5_due_fast (fast clone per fixtures.requires[1], Flow.type schedule hosting config.timeRelative, interval 60000) {type:time_relative, object:showcase_task, recordId:cCey-DFQU_dY7yd3}. N0: anonymous POST /automation/showcase_batch_reminders/trigger and /showcase_reassign_wizard/trigger -> 401 UNAUTHENTICATED, run count 1->1 for each
   automation.trigger-type-matrix · r10	A1	pass	HTTP-booted-server: urgent_task_alert runs before 0, after the mutations exactly 2: recordId kyy-DTdhm-p0szAD (created urgent) and 1uVksVze6EflHY2B (medium -> PATCH urgent); the later title-only save of 1uVk produced no run. The in_review -> done task AzKBJhwPUSCAtOCe fired showcase_task_completed and showcase_task_done_notify_owner once each, and its later title-only save fired nothing
   automation.trigger-type-matrix · r10	A2	pass	HTTP-booted-server: POST /automation/hooks/showcase_inbound_task_webhook/intake with t=‹now›,v1=HMAC('‹t›.‹body›') -> 202 {accepted:true,messageId:msg_faee6c49…}; task PH1J2di41Wv7v--- title/assignee/project = posted body (g5 hook …, g5.assignee@example.com, GOHSJxcxq3rVq9Fk), status todo. N1: body-only sha256=, wrong key and no header -> 401 INVALID_SIGNATURE, task total 14->14
   automation.trigger-type-matrix · r10	A3	pass	HTTP-booted-server (flag boot): showcase_scheduled_digest runs completed at 07:36:00.844, 07:37:00.843, 07:38:00.849, 07:39:00.852, 07:40:00.847Z, all trigger.type schedule, no manual help; sys_inbox_message project.digest rows for admin at 07:36:00.883, 07:37:00.879, 07:38:00.952, 07:39:00.923, 07:40:00.911Z
   automation.trigger-type-matrix · r10	A4	pass	browser-live + HTTP-booted-server: row action click captured POST /api/v1/actions/showcase_task/showcase_bulk_reassign {recordId:1uVk…} -> 200 {status:paused, runId:run_706616fa…, screen.nodeId:collect}; GET run -> paused; Submit captured POST /automation/showcase_reassign_wizard/runs/run_706616fa…/resume {inputs:{new_assignee}} -> 200; GET run -> completed, steps start/collect/apply success; dialog closed, new assignee shown in grid; no /trigger POST
   automation.durable-suspend-restart · r4	A0	pass	HTTP-booted-server: GET /data/sys_automation_run/:id: run_23600cd6 showcase_task_follow_up node hold (wait) paused vars 2224B ctx 2066B; run_d99d3142 showcase_project_closure node signoff (subflow) paused correlation subflow:run_5235b5b1 vars 2668B; run_5235b5b1 showcase_closure_signoff node ask_signoff (approval) paused correlation areq_ae4f2657 vars 2751B
   automation.durable-suspend-restart · r4	A1	pass	HTTP-booted-server (two spawned processes): boot A's process group SIGTERMed 07:41:47Z (port free, rows paused on disk); cold boot B over the same file DB ready 07:42:13Z; run_d99d3142 and run_5235b5b1 re-read paused with identical node, correlation, payload sizes and updated_at; run_23600cd6 still paused at the re-read; seeded data present (5 projects, task VEHdkRYC4kl2Xlq6)
   automation.durable-suspend-restart · r4	A2	pass	HTTP-booted-server (boot B): run_23600cd6 completed 07:42:41.164Z (remind step finished 54 s after boot A stopped, with no manual resume); sys_inbox_message task.followup "Follow up on: g5 durable follow-up" for admin at 07:42:41.193Z
   automation.durable-suspend-restart · r4	A3	pass	HTTP-booted-server (boot B): before: parent paused at signoff, child paused at ask_signoff; POST /approvals/requests/areq_ae4f2657…/approve -> 200 status approved; child completed output {decision:approved} (mark_approved ran); parent completed (signoff, notify_owner success); inbox row for usr_showcase_auditor_demo body "Closure sign-off decision for "Website Relaunch": approved."
   automation.durable-suspend-restart · r4	A4	pass	HTTP-booted-server: after resume GET /data/sys_automation_run/‹id› -> 404 RECORD_NOT_FOUND for run_23600cd6, run_d99d3142 and run_5235b5b1; each run's history reads completed. N0 held: no row vanished across the restart, and no completed run left a paused row
   automation.paused-run-trigger-record-masked · r5	A0	pass	HTTP-booted-server, admin AND member (g5.member, a member holding only the runtime set g5_run_reader with read on sys_automation_run + g5_echo): GET /data/sys_automation_run/run_12363338 -> 200 paused; variables_json record/$record/previous and context_json record/previous: g5_password and g5_token = SECRET_MASK '••••••••', the unset password field = null, internal g5_hidden key absent, name = written value; leak scan clean (plaintext pw old/new, token, internal value, 'secret:'). Arming: the primary DB row holds the plaintext password, a secret:sec_… handle and the internal value
   automation.paused-run-trigger-record-masked · r5	A1	pass	HTTP-booted-server, admin and member: GET /automation/qa_g5_mask_pause/runs/run_12363338 -> 200; variables.record, .previous and .$record: password and token masked, unset field null, g5_hidden absent, name kept; leak scan clean. negative[0] held for both personas, hot and after the cold boot (run_49e47d1a read by both on boot B)
   automation.paused-run-trigger-record-masked · r5	A3	pass	HTTP-booted-server (two spawned processes): run_49e47d1a paused on boot A; after SIGTERM and cold boot B over the same file, resumed on B -> 200; echo row "g5 echo c1": seen_name=g5 vault c1 renamed, seen_password/seen_token/seen_previous_password = mask, seen_hidden = ABSENT; paused row then 404 RECORD_NOT_FOUND
   automation.flow-registration-verdict-on-read · r2	A1	pass	HTTP-booted-server: GET /meta/flow/showcase_urgent_task_alert -> registration {state:registered}; scratch qa_g5_refused_flow (type api, no secret; saved 200 with OS_ALLOW_UNLINTED_METADATA_WRITES=1) -> registration {state:refused, reason.message "…its start node declares no config.secret…"} while item.status reads active; same after the cold boot
   automation.flow-registration-verdict-on-read · r2	A2	pass	HTTP-booted-server: GET /meta/flow -> {type, items[35], registrations{35}}; keys equal the item names; qa_g5_refused_flow refused, showcase_urgent_task_alert registered; states registered 34, refused 1
   automation.flow-run-semantics-v18 · r3	A1	pass	HTTP-booted-server: scratch flow qa_g5_stamp_user (record-after-update on showcase_private_note, writes current_user.id into body); member g5.member PATCH own note 36AVsIh_zeghkv7M title -> run_159e5070 completed, trigger.userId O1d2r4YA… (member); read-back body = O1d2r4YA3PmXvjeOal7kaGT3YgqY1mh6 = member id (not the admin id)
   automation.inbound-hook-timestamped-signature · r2	A1	pass	HTTP-booted-server: valid t=‹now›,v1= -> 202 {"accepted":true,"messageId":"msg_e5e0346d-…"}; run count 1 -> 2, task gB15o1qaX_opss-w created
   automation.inbound-hook-timestamped-signature · r2	A2	pass	HTTP-booted-server: body-only sha256=, t-400 s, unknown flow, wrong hook id (plus wrong key, no header) -> 401, all six bodies byte-identical {"success":false,"error":{"code":"INVALID_SIGNATURE","message":"Signature verification failed."}}; run count 1 -> 1, task total 14 -> 14
   automation.scheduled-work-opt-in · r2	A1	pass	HTTP-booted-server (default boot, flag unset): banner "1 flow declares a 'schedule' trigger but is NOT bound — disabled by deployment policy — … (OS_AUTOMATION_SCHEDULED_WORK_ENABLED is unset or not truthy) …: showcase_scheduled_digest" (and the same for time_relative: showcase_task_due_reminder); _status bound:false unboundCause deployment_policy for both; digest and due_reminder runs 0 at 07:43:07Z, 8 min after ready
   automation.scheduled-work-opt-in · r2	A2	pass	HTTP-booted-server (flag boot): no NOT bound line, "32 flow(s) 21 bound" (19 on the default boot), _status bound:true for both; showcase_scheduled_digest completed every 60 s from 07:36:00Z (five runs to 07:40:00Z, trigger.type schedule)
   automation.screen-flow-roundtrip · r7	A0	pass	browser-live+HTTP-booted-server: served /_console row-action click -> one POST /api/v1/actions/showcase_task/showcase_bulk_reassign {recordId,params:{}} -> 200 status:"paused", screen.nodeId "collect"; 0 POSTs to /automation/*/trigger; GET /automation/showcase_reassign_wizard/runs/:runId -> status paused, steps start:success,collect(screen):success (runs run_4b7b77c9…, run_7c303a37…, two fresh loads)
   automation.screen-flow-roundtrip · r7	A1	pass	HTTP-booted-server: GET …/runs/:runId/screen -> 200 {runId, screen:{nodeId:"collect", title:"New Assignee", fields:[{name:"new_assignee",type:"text",required:true}]}} (x2, and again on the cancel-parked run)
   automation.screen-flow-roundtrip · r7	A3	pass	browser-live+HTTP-booted-server: Submit sent POST …/runs/:runId/resume {inputs:{new_assignee:"qa-g6-r2-…@example.com"}} -> 200; dialog count 0; run re-read status completed, steps …,apply:success (x2)
   automation.screen-flow-roundtrip · r7	A4	pass	HTTP-booted-server: GET /api/v1/data/showcase_task/8rWxGV6PsV4sJ4EY -> record.assignee == the submitted value (r1 and r2 values each matched; authoritative read, not the grid)
   automation.screen-flow-roundtrip · r7	A5	pass	HTTP-booted-server: on a browser-launched paused run, POST resume {inputs:{}} -> 400 INVALID_SCREEN_INPUT "Screen field "new_assignee" is required"; also {} and {new_assignee:""} -> 400; run re-read paused (steps end at collect); later valid resume -> 200, run completed, task assignee = value (x2). N0 holds
   automation.screen-flow-roundtrip · r7	A6	pass	browser-live+HTTP-booted-server: dialog buttons [Cancel, Submit, Close]; Cancel -> 0 dialogs, only the launch POST sent; run re-read paused at collect, task assignee unchanged; /screen 200 same contract; valid resume -> 200, run completed, task assignee = value (x2). N1 holds
   automation.setup-packaged-automation-board · r10	A0	fail	HTTP-booted-server+browser-live: GET /api/v1/meta/app?id=setup group_apps holds nav_packages, nav_packaged_actions (type component, componentRef automation:packaged-actions, label "Packaged Actions"), nav_marketplace_installed — 0 occurrences of nav_packaged_automation; sidebar shows "Packaged Actions" (0 "Packaged Automation" links), click lands /component/automation/packaged-actions; typed retired URL …/component/automation/packaged redirects to /studio/com.example.showcase/metadata/flow (x2). Stale spec: #28785 (ADR-0146 D11/D4)
   automation.setup-packaged-automation-board · r10	A1	fail	browser-live+HTTP-booted-server: the served Setup page has 0 packaged-flow-* rows (flows half retired by #28785); actions section renders (140 rows). Replacement: Studio › Automations rail lists 32 flows = GET /automation/_status 32, only showcase_approver_bindings enabled:false, 2 schedule flows unboundCause deployment_policy. Stale spec
   automation.setup-packaged-automation-board · r10	A2	fail	browser-live+HTTP-booted-server: no flow switch on the Setup page (retired). Replacement does what the clause wants: Studio auto-ledger-switch on showcase_urgent_task_alert -> POST …/toggle 200 enabled:false; after full reload aria-checked=false, data-reading off, "Off for this deployment"; sys_metadata_activation row active:false; _status enabled:false; flipped back -> active:true (x2). Stale spec
   automation.setup-packaged-automation-board · r10	A3	fail	browser-live: Packaged actions page has 0 Clone controls and 0 "New similar" (the actions half the charter protects HOLDS); "every flow row carries Clone" has no rows to read — flow copy moved to Studio's ⋯ › New similar (auto-item-new-similar). Stale spec
   automation.setup-packaged-automation-board · r10	A4	pass	browser-live: with a flow clone present (qa_g6_s1_clone in com.example.showcase.custom), the served Packaged actions page has no customized/diff/cloned-from/drift/lineage text and 0 lineage/drift/diff/cloned testids; Studio's view of the clone shows none either (x2)
   automation.setup-packaged-automation-board · r10	A5	fail	browser-live+HTTP-booted-server: no flow toggle on the Setup page (retired). Admin cell on the replacement: Studio switch on showcase_notify_owner -> 409 DELETE_RESTRICTED, sentence naming showcase_task_done_notify_owner rendered verbatim in data-testid=studio-refusal — but NOT role=alert (getByRole('alert') = 0) and under the tab bar, not beside the switch. Member cell: Setup 403 page, Studio "You need Studio access"; API toggle -> 403 "Enabling or disabling an automation flow requires the manage_metadata capability." (exact). Stale spec
   automation.setup-packaged-automation-board · r10	A6	fail	browser-live: the page's clone dialog is retired. Replacement does what the clause wants: Studio New similar opens prefilled (…copy / "… (copy)"); submit disabled with name empty, label empty, both empty; Bad-Name! -> POST …/clone 400 VALIDATION_FAILED, role=alert auto-new-similar-refusal "Invalid flow definition: name: Invalid string: must match pattern /^[a-z][a-z0-9_]*$/", dialog stays open; legal clone -> 200, notice "Saved the copy … in package com.example.showcase.custom. The copy is switched off …" (x2). Stale spec
   automation.setup-packaged-automation-board · r10	A7	pass	browser-live+HTTP-booted-server (posture recorded, fresh-context member minted invite-first, form sign-in, sys_member role member): /component/automation/packaged-actions renders "You don't have access to this app" (GET /meta/app/com.objectstack.setup 403 PERMISSION_DENIED), 0 switches; Studio flow URL renders "You need Studio access to open Studio"; sys_metadata_activation read 403; GET /automation/_status 200; GET /meta/action 200; action toggle POST /actions/_activation/showcase_task/showcase_archive_task -> 403 manage_metadata sentence
   automation.flow-authoring-refusals-v18 · r5	A1	pass	cli-spawned (os validate --json, scratch app, one arm per run): every arm exit 1 — value token {record.owner} and text token {record.name} -> expression-invalid at flows[32].nodes[1].config.fields.owner / .config.title (names CEL envelope / (double-brace record.name template)); record.x in autolaunched no-entrance flow and user.id -> expression-invalid at .config.recipients (names current_user.id); $x screen field, outputVariable:42, notify bogusKey, dangling edge (edges.2.target), repeated edge (edges.2), job retryPolicy.maxRetry (did-you-mean maxRetries), try_catch retry.maxRetry, anonymous armed-rateLimit endpoint -> runAs:'system' flow (apis.3.authRequired, ADR-0138 D2b) -> STACK_SCHEMA_INVALID with the contract sentence at the located path. os build corroborates (bogusKey exit 2, record.x exit 1 "Author-time rules failed … expression-invalid", control exit 0). N0 holds: no arm validated clean
   automation.flow-authoring-refusals-v18 · r5	A2	pass	HTTP-booted-server (boot of the scratch app declaring authRequired:false qa_anon_api -> tenant name qa_anon_target): PUT /api/v1/meta/flow -> 422 INVALID_METADATA with located issues for value token, text token, $x, outputVariable 42, bogusKey, dangling edge, repeated edge, try_catch retry.maxRetry, and qa_anon_target runAs:system (issue path "runAs"); each GET -> 404. record.x and user.id arms -> 200, GET 200. PUT /meta/job and /meta/api -> 403 NOT_CREATABLE (x2, second pass with fresh names)
   automation.flow-authoring-refusals-v18 · r5	A3	pass	cli-spawned+HTTP-booted-server: os validate exit 0, 0 warnings on the planted flow, for CEL envelope {dialect:cel,source:record.owner}, notify title "Case (double-brace record.name template)", get_record filter {owner_id:'{current_user_id}'}; PUT each -> 200, GET 200 (x2)

   cli

   item · rev	clause	verdict	layer · text evidence
   cli.scaffold-first-run · r7	A0	pass	cli-spawned + HTTP-booted-server (PRE-RELEASE variant): 74 subject tarballs (pnpm pack, scripts ignored) on local verdaccio 6.11.1 :41671 with no uplink for @objectstack/* / create-objectstack (GET /create-objectstack 404 before publish); repo bin create-objectstack.js qa-first-run -t blank --skip-skills exit 0, pnpm install ran, no "Dependency installation failed", node_modules/@objectstack/cli 18.2.0; lockfile integrity of all 60 @objectstack/* entries == local tarball sha512 (60/60); npm run validate 0, npm run build 0; npx os start --artifact ./dist/objectstack.json --port 41672 health 200 {"status":"ok","version":"18.2.0"} on try 7 of 30, /api/v1/ready 200 {"status":"ready","state":"running"}
   cli.scaffold-first-run · r7	A1	pass	cli-spawned: variants=[blank] (the one template in template-registry.ts): validate exit 0, build exit 0 → dist/objectstack.json 2.5 KB, over the subject's runtime (integrity-proven install above)
   cli.scaffold-first-run · r7	A2	pass	HTTP-booted-server: os start --artifact of the scaffold booted healthy, server.log has no protocol refusal line ("✓ Server is ready", Mode production); scaffold pins 18.2.0 == candidate tarballs, so no RC-window skew applies
   cli.scaffold-first-run · r7	A3	pass	cli-spawned: 18-line scaffold without --skip-skills (qa-skills): "Installing AI skills … from @objectstack/skills@18.2.0", "Found 11 skills in node_modules"; .claude/skills SKILL.md dirs (sort -u) == node_modules/@objectstack/skills/dist/skills SKILL.md dirs, 11 = 11, SET-EQUAL, 0 symlinks, no .agents left, no dogfood-verification. GitHub-path probe on a fresh skills-probe scaffold: npx -y skills add ‹checkout›/skills --all --copy (skills 1.7.4) exit 0, installed 11 == curated skills/ 11, SET-EQUAL; repo-root --list "Found 11 skills", dogfood-verification hits 0. N0 held
   cli.dev-automigrate-policy · r8	A0	pass	cli-spawned (os dev, scratch app, config-declared sqlite qa_main + datasourceMapping [{default:true, datasource:'qa_main'}], autoMigrate 'safe'): three drifts staged (nn_field storage.notNull dropped = needs_confirm; doomed removed = destructive; legacy qaam_item_uq_field_unique planted on the stopped file = safe). Safe boot 1 log: exactly one [schema-drift] auto-reconciled replace_unique_index on qaam_item.uq_field, nn_field and doomed still warned; qaam_item dump 6793ff21f5b775e7 → 1017b7100dcac006 (legacy index gone, tenant-scoped index kept). Safe boot 2: no auto-reconciled line, dump unchanged (converged)
   cli.dev-automigrate-policy · r8	A1	pass	cli-spawned: three boots over the staged drift — autoMigrate 'off', key omitted, 'off' again: each printed the same three [schema-drift] warnings (line sets diff-identical across boots), 0 auto-reconciled lines; full-schema dump 33d4895c3ad4976b and qaam_item dump 6793ff21f5b775e7 byte-identical before boot 1 and after boot 3 (legacy index still present). N1 held
   cli.dev-automigrate-policy · r8	A2	pass	cli-spawned (NODE_ENV=production os serve objectstack.config.ts, OS_AUTH_SECRET + 64-hex OS_SECRET_KEY, legacy index re-planted, autoMigrate 'safe' declared): log carries verbatim [schema-drift] autoMigrate='safe' is ignored under NODE_ENV=production — schema is never auto-altered in production. Run 'os migrate' deliberately., the three drift warnings, 0 auto-reconciled; "✓ Server is ready", Mode production; dump before == after (legacy index kept)
   cli.dev-automigrate-policy · r8	A3	pass	cli-spawned: destructive orphan doomed and needs_confirm nn_field warned on safe boot 1 and again on safe boot 2, never applied (dump still carries doomed text and nn_field text not null). schemaMode gate, both sides on one side datasource qa_side (autoMigrate 'safe', legacy qaam_side_code_unique planted): schemaMode 'external' boot → no drift line for qaam_side, side.db dump unchanged (legacy index kept); same config with schemaMode 'managed' → [schema-drift] auto-reconciled replace_unique_index on qaam_side.code, index gone. N0 held
   cli.migrate-meta-write-stored-and-all · r3	A3	pass	cli-spawned: minimal stack (one object, no plugins/jobs/agents) os migrate meta --from 17: "427 manual change(s) require your judgment" + one line "21 more manual change(s) not listed: their surfaces are absent from this stack (run with --all to list them)."; --all: "21 manual change(s) whose surfaces are absent from this stack (listed by --all)" with 21 "absent: nothing is declared under …" lines (0 in the default run); --json todos 448, absentTodos 21
   cli.migrate-meta-write-stored-and-all · r3	A4	pass	cli-spawned: scratch DB (booted once, stopped) with two sys_metadata flow rows — get_record filter {id:'{record.id}'} and CEL control {dialect:'cel',source:'record.id'}; os migrate meta --stored --json: scanned 2, canonical 1, skipped 1, row qa_meta_refused todos[0] conversionId flow-value-slot-template-dialect-refused path flows[0].nodes[1].config.filter.id; text mode prints TODO flow-value-slot-template-dialect-refused: "{record.id}" left as stored at flows[0].nodes[1].config.filter.id … and "☐ TODO: 1 site(s) in 1 row(s) are left as stored"; DB sha unchanged (preview)
   cli.validate-v18-authoring-verdicts · r2	A1	pass	cli-spawned (scratch showcase copy, one plant per arm, reverted between arms): seed created_atx → validate exit 1 "defineSeed('showcase_task'): unknown field(s) … records[0]: created_atx … Did you mean 'created_at'?", build exit 2; record["typo"] → rule expression-invalid at objects[16].validations[1].condition "unknown field typo on showcase_project", build exit 1; datasource twice → STACK_CROSS_REFERENCE_INVALID "declared twice … stack.datasources[0] and stack.datasources[1]", build 2; cbp master_detail required:false → ObjectSchema.create('showcase_invoice_line') field invoice contradiction, build 2; object-grid defaultFilters → component-props-invalid at pages[19].regions[1].components[2].properties.defaultFilters, build 1; email {apiKey:'x'} → STACK_SCHEMA_INVALID "email: Unrecognized key(s) … apiKey" prescribing OS_EMAIL_API_KEY, build 2. Every arm: no "Validation passed", no dist/objectstack.json
   cli.validate-v18-authoring-verdicts · r2	A2	pass	cli-spawned: controls — seed key created_at (stock), record.name == "record.typo" literal in the same rule, cbp master_detail required (stock), email: { provider: 'log' }: each validate --json exit 0 valid:true, text "Validation passed", build exit 0 with artifact
   cli.verify-derived-proof-families · r3	A1	pass	cli-spawned (scratch showcase copy): os verify --X --json each exit 0, top-level hardFailures 0, family key present: referenceShapes (24 fields, 8 proven, 0 holes, 16 inconclusive), deleteBehaviors (9 proven, 0 contradicted), uniqueScopes (1 proven), parentWriteReach (2 objects proven, 180 held), dashboardWindows (17 widgets proven, 273 probes held) — summary.hardFailures 0 in all five

   ---

   Generated by Claude Code
4. objectstack-fleet commented on Nov 2, 2026

   

   objectstack-fleetBot

   on Nov 2, 2026

   ContributorAuthor

   More actions

   domain:devx#1 · os-warren · session_01JQwxbZx1yZXiTrFXKbSdFZ

   Close-out for this record (SKILL §5).

   One item failed, automation.setup-packaged-automation-board, on six clauses that pin the Setup › Packaged automation page #28785 retired: a stale spec whose re-point #28579 already carries (run evidence posted there). Six product defects observed beside the clauses are extracted; each passed the pm-dispatch filing gate (class ①). A seventh, os migrate meta --stored --json attesting clean over an absent database, is the same family as the open #28947 and was added there as a comment. The two PARTIAL items' blocked clauses and the checklist-accuracy lines are parked on the wave anchor.

   Extracted: #28579 · automation.setup-packaged-automation-board · acceptance[0] acceptance[1] acceptance[2] acceptance[3] acceptance[5] acceptance[6] (stale spec, existing re-point card)
   Extracted: #29130 · access-security.share-link-landing-page · observed beside the clauses
   Extracted: #29133 · automation.screen-flow-roundtrip · observed beside the clauses
   Extracted: #29134 · cli.scaffold-first-run · observed beside the clauses
   Extracted: #29135 · ai.mcp-http-surface · observed beside the clauses
   Extracted: #29137 · api-backend.bulk-write-contract · observed beside the clauses
   Extracted: #29139 · automation.setup-packaged-automation-board · acceptance[5]
   Extracted: #28947 · cli.migrate-meta-write-stored-and-all · observed beside the clauses (comment on the existing card)
   Parked-on: #27377 · api-backend.action-activation-door-contract · acceptance[4] blocked (fixture, knownGaps[3]: metadata-plane 503 arm has no stock door)
   Parked-on: #27377 · approvals.email-action-token-door · acceptance[0] acceptance[1] acceptance[3] acceptance[4] acceptance[5] blocked (fixture, knownGaps[0]: no production token producer)
   Parked-on: #27377 · 25 checklist-accuracy findings and 5 pin-promotion candidates (anchor comment)
   

   Observations below the filing gate stay in this record only: the bounded-probe warning's remedy "scope by parent_object" does not end the truncation when one parent object alone holds more than 2000 rows (log advice only); an admin may attach a file to a record of a read-only federated object (the join row is in the primary DB); stock showcase notify flows end failed on a task with no assignee (the notify contract working on optional data); MCP get_record with an empty id says "Record not found" (wording only).

   No security finding.

   ---

   Generated by Claude Code