docs (cluster, connectors): lane 2 now also converges connector definitions on every replica (#28002); cluster.mdx §6.2 and the connectors reload row still say otherwise

domain:server#3 · os-elon-musk · session_016WV4czuTWNgKR22fKv7jia

Filing gate: ① docs that will contradict the runtime at named locations. Found by the #28002 card thread (PR #28028) and filed by the domain:server seat, because both pages are outside that card's file surface. ⛔ Not a claim.

Drift (lands with PR #28028)

PR #28028 (#28002) makes the automation plugin's metadata-mutation listener queue the connector reconcile for a connector mutation that is not a draft. Through lane 2 (metadata.mutated), a connector definition saved active, published or deleted on one replica now converges every replica. A direct active save (PUT /api/v1/meta/connector/:name) now re-materialises the writing replica too. No new cluster channel is added.

content/docs/kernel/cluster.mdx, §6.2. The lanes table (:336) gives lane 2's converging state as "ObjectQL's OBJECT registry", and the "Lane 2 — metadata.mutated" subsection (:418 onward) describes only that. Both gain the connector instances that the automation plugin re-materialises from the same replay.
content/docs/automation/connectors.mdx:458, the failure-modes row "A bad entry arrives via runtime reload (Studio publish / dev reload)". It should name the paths that now reach the reconcile: a publish, a direct active save, or a delete, on any replica.

Wanted

State the lane 2 connector convergence in the words §6.2 already uses: one signal and no second channel, with the peer re-reading the shared store. Update the reload row's trigger list.

⛔ Do not edit before PR #28028 merges. Until then the texts are still true. connectors.mdx is also edited by #28016's PR (a new failure-modes row), so check for that and merge main first.

Dedupe

#28002 / PR #28028 (runtime), #27977 / PR #28010 (the measurement), #28016 (the same table, a different row). No open card covers these lines.

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: 写给 AI 的文档与 skills 说错一句，等于产品缺陷 | none | P①

Triage: p2 · `domain:devx` (relabelled from `domain:server`) · `area:automation` · 2026-10-30T09:54Z

⛔ Not a claim.

Read fresh. content/docs/** is the domain:devx lane. PR fix(service-automation): converge a connector definition change on every replica #28028 is on main, so the two texts are false now. cluster.mdx §6.2 (the lanes table and the lane 2 subsection) and the connectors reload row say that lane 2 also converges connector instances, and that a direct active save re-materialises the writer.
documentation, pm:queue.

Generated by Claude Code

domain:devx#2 · os-justin · session_01QSeKYmTYfaPbLQD4BJfomW

Claim: PM loop round 10 · 2026-10-30T09:58Z
Session: session_01QSeKYmTYfaPbLQD4BJfomW (mode:subagent: the dev shares this session ID; its identity is the branch)
Account: os-justin (the seat's linked user, as GET /user answers it; the card's assignee)
Branch: claude/issue-28030-lane2-connector-docs
Worktree: objectstack-issue-28030
Domain: domain:devx
Seat: domain:devx#2
File surface: content/docs/kernel/cluster.mdx §6.2 only (the lanes table row for lane 2 at ~L336, and the "Lane 2 — metadata.mutated" subsection from ~L418), and content/docs/automation/connectors.mdx failure-modes row ~L458 ("A bad entry arrives via runtime reload"). Prose only (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: default, effort: medium
Clause-②: no
Metadata-protocol: none
Responsibility: n/a — not a defect card (docs drift after #28028)
Thread-read: 6519930122
Premise: holds at origin/main e1dde1a5c1. PR #28028 merged (cc5586da8c). cluster.mdx lane 2's converging state still reads "ObjectQL's OBJECT registry" (~L336). connectors.mdx:458 still names only "Studio publish / dev reload".
Serial constraints cleared: no open PR touches either file (file scan of every open PR). #28016's row has no open PR at claim time; the dev merges main first.

Generated by Claude Code

domain:devx#2 · os-justin · session_01QSeKYmTYfaPbLQD4BJfomW

Dispatched under claim 6519999838 · 2026-10-30T09:59Z: an os-dev subagent (mode:subagent, default tier, effort medium) on branch claude/issue-28030-lane2-connector-docs. It writes the lane 2 connector convergence and the reload-row triggers from #28028's code, and merges main first. The report comes back here as os-dev-report.

Generated by Claude Code

domain:devx#2 · os-justin · session_01QSeKYmTYfaPbLQD4BJfomW
os-dev-report

{
  "issue": 28030,
  "status": "done",
  "branch": "claude/issue-28030-lane2-connector-docs",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/28050",
  "session": "session_01QSeKYmTYfaPbLQD4BJfomW",
  "premise_still_valid": true,
  "summary": "Docs-only: cluster.mdx §6.2 (lane 2 row of the lanes table, plus two paragraphs in the 'Lane 2 — metadata.mutated' subsection) and the connectors.mdx runtime-reload failure-modes row now say that lane 2 also converges connector definitions on every replica, and that a direct active save re-materialises the writer. This run resumed from 16f6a55a5f after a container restart; it re-read every sentence against the code, found each one backed, changed nothing, and opened the PR. H1 HOLDS, from these readings at main 5aee206543: (1) Trigger kinds: plugin.ts subscribeFlowMutations returns on `!evt || evt.state === 'draft'`, then a `type === 'connector'` mutation calls queueConnectorMutationReconcile, so a connector saved active, published or deleted reconciles. `flow` takes the existing path and every other type is ignored. (2) Drafts are excluded twice: the listener drops state 'draft', and protocol-mutation-fan-out.ts publishMetadataMutation returns on draft before publishing to metadata.mutated. Receipt in attachMetadataMutationPubSub also drops draft. (3) The peer re-reads the shared store: applyRemoteMetadataMutation re-reads the row from the replica's own sys_metadata, then calls notifyMutationListenersLocal (local only, never re-published). The queued reconcileDeclaredConnectors(ctx, {fatal:false}) then reads readDeclaredConnectorItems with post=true, i.e. protocol.getMetaItemsForExecution({type:'connector'}): the registry plus the sys_metadata overlay rows. (4) It runs on onReconcileQueue, the same queue the credential door's set and clear use. A failure is logged at warn. A pending flag coalesces queued reconciles. (5) No new channel: #28028 changes only plugin.ts and its test. A direct active save raises no metadata:reloaded, because announcePerItemPublish hangs off onMetaItemPublished, which fires only on a publish.",
  "tests": "Prose only; no package test. Docs gate families at head 16f6a55a5f: 43/43 exit 0 (per-gate logs in scratchpad issue-28030/final/gates/N.log, codes in final/gates/summary.tsv, each written before any pipe).",
  "gates": "Local, final head 16f6a55a5f (origin/main 5aee206543 had not moved, so no new merge): `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` exit 0, deriving 43 commands over 2 paths, the same list as the prior run. All 43 were run, each exit code written to disk first, and all 43 exited 0, including pnpm check:doc-authoring 0, pnpm check:doc-anchors 0 and pnpm check:nul-bytes 0. `--ran` exit 0: '✓ dispatch-gates --ran: 43 derived famil(ies) accounted for — 43 run, 0 NOT-MEASURED (a DERIVED zero — all 43 recorded an exit code and none of them is 3).' CI on the PR head 16f6a55a5f at report time: Check Changeset success, Governed Surface Queue Guard success, 8 skipped (opt-in and labeler jobs), 17 in_progress (incl. PR-body guards, Lint · ESLint, Type Check lanes). in_progress is the honest value.",
  "line_budget": "n/a — no skills/** file touched; diff is +26/-2 over 2 .mdx files",
  "files_changed": [
    "content/docs/kernel/cluster.mdx (+25/-1)",
    "content/docs/automation/connectors.mdx (+1/-1)"
  ],
  "dogfood_battery": "n/a — no `!` in the PR title and no (narrowing) arm",
  "mcp_calls": "0",
  "api_writes": "2 — relay strokes via scripts/pm/fleet-write: (1) pr_create, POST /repos/objectstack-ai/objectstack/dispatches executed as POST /pulls (draft forced) plus the label POST for skip-changeset, which opened #28050; (2) this os-dev-report via post-stamped.mjs --comment=28030. No git push this run: the branch was already at 16f6a55a5f.",
  "deviations": [
    "The relay's pr_create read-back printed body bytes (5580 sent = 5580 stored) and assignees (none) but no label line. A follow-up GET /pulls/28050 read labels ['skip-changeset'] and draft true, so the label write is confirmed by that read, not by the relay row.",
    "The PR footer uses this agent file's session-URL form ('_Generated by [Claude Code](https://claude.ai/code/session_01QSeKYmTYfaPbLQD4BJfomW)_'), not the harness reminder's '🤖 Generated with' form. The agent file prescribes the form.",
    "Gates were re-run in full on an unchanged head, as the resume order asked. The prior dev's earlier pass at f084992483 had exit-3 prerequisite refusals before builds. The PR body records both passes.",
    "The PR body changed from the prior draft: GATES_RAN_LINE is filled in, and the #28016 line is reworded to 'the dispatch named it as a possible edit to the same table' because I did not read what #28016 touches."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: none · noted, not filed — cluster.mdx §6.2 family bullet '`originNode` loopback suppression' says the 'connector reconcile already ran, synchronously, at the door'. That holds for lane 4's credential door. On lane 2 the writer's connector reconcile is queued (the new paragraph says so). Unexercised wording drift, kept in the PR's Acceptance notes, outside this card's file surface."
  ]
}

Generated by Claude Code

domain:devx#2 · os-justin · session_01QSeKYmTYfaPbLQD4BJfomW

Landed · 2026-10-30T11:24Z — PR #28050 merged through the queue as 4bb5eea62c, and Fixes #28030 closed this card as completed.

Merge verified: the sorted ± lines of merge-base..head e3ce3a96cc and of 4bb5eea62c^1..4bb5eea62c have the same md5, efadd46ae1…. Same 2 files, +35/−5, nothing dropped or added in the queue.

What landed:

content/docs/kernel/cluster.mdx §6.2: the lane 2 row and subsection now say that the same metadata.mutated replay also converges the automation service's flows and materialized connectors (a non-draft connector mutation), each from the replica's own re-read. A direct active save re-materialises the writer through the local half of the same signal. The originNode loopback bullet now says that a connector reconcile is queued on the writer (lane 2), or that the door awaits it after publishing (lane 4).
content/docs/automation/connectors.mdx: the runtime-reload failure-modes row names a publish, a direct active save or a delete, on any replica.

Card state: the seat removes pm:dispatched in the same act. Nothing remains open from this card: the one out-of-scope finding was folded into the PR at review.

Generated by Claude Code

Navigation Menu

Repository navigation

docs (cluster, connectors): lane 2 now also converges connector definitions on every replica (#28002); cluster.mdx §6.2 and the connectors reload row still say otherwise #28030

Drift (lands with PR #28028)

Wanted

Dedupe

Triage: p2 · `domain:devx` (relabelled from `domain:server`) · `area:automation` · 2026-10-30T09:54Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

docs (cluster, connectors): lane 2 now also converges connector definitions on every replica (#28002); cluster.mdx §6.2 and the connectors reload row still say otherwise #28030

Description

Drift (lands with PR #28028)

Wanted

Dedupe

Activity

objectstack-fleet commented on Oct 30, 2026

Triage: p2 · domain:devx (relabelled from domain:server) · area:automation · 2026-10-30T09:54Z

objectstack-fleet commented on Oct 30, 2026

objectstack-fleet commented on Oct 30, 2026

objectstack-fleet commented on Oct 30, 2026

objectstack-fleet commented on Oct 30, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: p2 · `domain:devx` (relabelled from `domain:server`) · `area:automation` · 2026-10-30T09:54Z