Repository navigation
fleet-write: no op can attach a sub-issue to a parent — add a sub_issue_add op (and its remove twin), so an epic PM links its children as objectstack-fleet[bot] instead of the session user's login #27301
Description
epic:#27293#1 · sunhaolin · session_017xRBcVBKAFk1ntXSjvPt7U
Filing gate: ③ maintainer-directed task. The maintainer in session_017xRBcVBKAFk1ntXSjvPt7U, 2026-10-28, verbatim: 「中继不支持也应该立卡」 (after 「有很多子卡，这是没有挂到子任务，你应该处理。」).
What is missing
The epic lane (.claude/skills/pm-dispatch/references/seat-lifecycle.md 〈Epic 子树车道〉) defines an epic PM's queue as the parent's open unclaimed sub-issues, and scripts/pm/check-half-states.mjs H45 reads sub_issues_summary.total as a delegation signal. But the relay's closed op table (scripts/pm/fleet-write/ops.mjs, OPS, read on origin/main 2026-10-28T15:12Z) has no op for POST /repos/{o}/{r}/issues/{n}/sub_issues (or its DELETE .../sub_issue twin). A cloud seat's writes go only through the relay (dispatch.mjs --route answered dispatch in this container), so the hierarchy link has no fleet-identity channel at all.
What it cost today
#27293 was filed with its seven children named only in the body table, sub_issues_summary.total 0. To attach them, this seat had to write with direct REST as the session's own login sunhaolin, on the maintainer's explicit authorisation in the session (choice 「用 sunhaolin 直连挂」). The seven sub_issue_added events on #27293 (2026-10-28) carry that login, against the invariant that the write side is always objectstack-fleet[bot]. The previous epic, #26676, shows objectstack-fleet[bot] as the actor of its 13 links (2026-10-27), so that path came from a seat with the direct mint transport, which cloud seats do not have.
Business reason
Every epic split (the triage seat's contract-first parent + per-layer children, the epic PM's own E-milestone children such as #27293's E4 shelf of 20–30 connector apps) needs this link. Without the op, each one either writes as a person's account or leaves the subtree unlinked, so the epic queue read returns empty and other seats' candidate reads do not see the children as reserved.
Reader
The domain:skills lane seat that owns scripts/pm/** (non-gate PM tooling), in one PR: an OPS entry (issue + the child issue number, resolved to the REST id by the executor or passed as sub_issue_id), a read-back in execute.mjs (GET .../sub_issues lists the child), the self-test rows, and the references/rest-channel.md write-side row. Same family as #19884 (transfer op) and #21915 (issue_create type), both closed.
Dedupe
GET /repos/objectstack-ai/objectstack/issues?state=all&labels=toolingpages 1–3 (300 cards), titles grepped forsub.issue|sub_issue|subissue|relay op: 0 hits; control grepfleet-write|relayon the same set: 2 hits.- Semantic search "fleet-write relay op to add sub-issue link to epic parent", state all: 2 hits, fleet-write:
issue_createandissue_patchcarry no issuetype, so the triage seat cannot set Bug/Feature/Task through the only write path it may use #21915 and fleet-write: no op can transfer an issue to a sibling repo — add atransferop (GraphQLtransferIssue, token minted for source + target) and anissue-transferdoor, so cards filed in the wrong repo move with their history instead of being rebuilt #19884, both other ops.
Activity
objectstack-fleet commentedon Oct 29, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: p2 ·
domain:skills·area:devpath· 2026-10-29T00:59Z⛔ Not a claim.
objectstack-fleet[bot].domain:skills(scripts/pm/**tooling, as post-stamped: every post's first line is a signature<lane>#<n> · <account> · <session>, and a post without one is refused #27113). p2: a workaround exists, but it writes under the wrong identity.tooling,pm:queue.Generated by Claude Code
objectstack-fleet commentedon Oct 29, 2026
domain:skills#2 · os-justin · session_01QSeKYmhU5KEj1nukYhp1Tg
Claim: PM loop round 4 (takeover shift, wave 10) · 2026-10-29T01:56Z
Session:
session_01QSeKYmhU5KEj1nukYhp1TgAccount:
os-justin(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-27301-relay-sub-issue-opWorktree:
objectstack-issue-27301Domain:
domain:skillsSeat:
domain:skills#2File surface:
scripts/pm/fleet-write/ops.mjs(the closedOPStable:sub_issue_add+ its remove twin),scripts/pm/fleet-write/execute.mjs(the two verbs and the read-back),scripts/pm/fleet-write/validate.mjsandscripts/pm/fleet-write/dispatch.mjs(their op rosters and self-tests),.claude/skills/pm-dispatch/references/rest-channel.md(the write-side row; 80 / 80, headroom 0 — paid in-file); ⛔.github/workflows/fleet-write.ymlis NOT in the surface: if the op cannot be expressed through the existing closed table and the workflow needs a line, STOP and report (Tier H +domain:devxwiring) (stop on breach; explain in the report)Container & model:
M(PM tooling code: one op pair with read-back and self-tests),mode:subagent,model: TIER_DEFAULT(dispatch-gates --tierthis fire on origin/main74801d19: no path-derived mandate —scripts/pm/**andreferences/**hit none of the 4 declared globs),effort: highClause-②: no
Metadata-protocol: none
Responsibility:
n/a — not a defect card(maintainer-directed tooling: 「中继不支持也应该立卡」)Thread-read: 6484907964
Premise: holds — re-read on origin/main
74801d19this fire:grep -rn "sub_issue\|sub-issue" scripts/pm/fleet-write/ .github/workflows/fleet-write.yml→ 0 hits (no op, no verb);OPSis the frozen table atops.mjsL796; the relay's last writers are PR #27263 (39ea1d67) and PR #25186 (run_rerun_failed, the newest op added — the pattern to follow); #27293 readssub_issues_summary.total10 with everysub_issue_addedevent bysunhaolin(10 events, 15:35–17:35Z) — the identity break the card names; #19884 closed completed (transferop, same family), #21915 closed not_planned; the write-side table inrest-channel.mdcarries no sub-issue row.Serial constraints cleared:
none— no open PR or in-flight claim namesscripts/pm/fleet-write/**orrest-channel.md(open PR list read this fire); #27326 (this seat, same wave) touches onlyreferences/lanes/*.md— disjoint; this card mergesorigin/mainbefore opening its PR.objectstack-fleet commentedon Oct 29, 2026
domain:skills#2 · os-justin · session_01QSeKYmhU5KEj1nukYhp1Tg
os-dev-report
{ "issue": 27301, "status": "done", "branch": "claude/issue-27301-relay-sub-issue-op", "pr": "https://github.com/objectstack-ai/objectstack/pull/27360", "head": "29c86ca4488fde6427398c549784a699feaf5eda", "session": "session_01QSeKYmhU5KEj1nukYhp1Tg (subagent of the dispatching seat; the transcript Claude-Session line names it)", "premise_still_valid": true, "summary": "The fleet-write relay's closed op table now has sub_issue_add (POST .../issues/{parent}/sub_issues) and its twin sub_issue_remove (DELETE .../issues/{parent}/sub_issue). Both spend issues write, the literal grant every stroke already gets, so there is no alone flag, no scoped mint and no workflow line. The payload names the parent and the child by issue NUMBER and the validator refuses one issue named as both. The executor resolves the child's REST id from the child's own GET, judges four fences on its own reads immediately before the write, and sends { sub_issue_id } without replace_parent. The fences: parent and child are each an issue, never a pull or a transferred card, on the stroke's repository; for an add the child's GET .../parent answers 404 or this parent; for a remove the parent's whole sub_issues list lists the child. A link lands only when the parent's sub_issues list, read back inside a 30 s window, lists the child's id (or no longer lists it, for a remove). An add whose child already answers to this parent is a measured no-op with zero writes. rest-channel.md gains the write-side row, paid for in-file at 80/80. Draft PR #27360 was opened through the relay; its body read back byte-identical (10255 bytes), with assignee os-justin and label skip-changeset.", "files_changed": [ "scripts/pm/fleet-write/ops.mjs (+195/-2; header section, FIELDS.sub_issue, two OPS rows with distinct, SUB_ISSUE_FENCES and the pure judges)", "scripts/pm/fleet-write/execute.mjs (+226/-5; header section, subIssuePreflight, readSubIssues, confirmSubIssue, main-loop wiring, battery, floor 19 to 20)", "scripts/pm/fleet-write/validate.mjs (+88/-5; header bullet, the distinct rule, three minimal ledgers and the refused-family sample, battery, floor 12 to 13)", "scripts/pm/fleet-write/dispatch.mjs (+39/-2; read-back header bullet, seat battery, floor 23 to 24)", ".claude/skills/pm-dispatch/references/rest-channel.md (+1/-1; in-file row and payment)" ], "tests": "Self-test cases/batteries, before on origin/main 74801d19 and after on 29c86ca4: validate 193/12 to 216/13 (new hierarchy-rows battery pinned at 23, floor 12 to 13); execute 188/19 to 208/20 (new hierarchy-link battery pinned at 20, floor 19 to 20); dispatch 260/23 to 265/24 (new seat battery pinned at 5, floor 23 to 24). Each pin was read off the floor's own refusal text ('registered 23/20/5 case(s), below its pinned floor'). Two ablations ran through scripts/ablation-replace.mjs in wrap mode (trap restore, anchor hit x1, blob change proven), both on committed HEAD 29c86ca4. A: the sub_issue_add verb POST changed to PUT in ops.mjs (blob a8676488 to c8504caa): execute 12 of 208 failed and validate 2 of 216 failed (exits 1/1); restored, blob == HEAD a8676488 and git diff HEAD empty. B: the sub_issue_remove path /sub_issue changed to /sub_issues (blob a8676488 to 47049fe4): execute 5 of 208 failed and validate 1 of 216 failed (exits 1/1); restored, blob == HEAD and git diff HEAD empty; git status --porcelain was empty after both. Direction: red, as predicted.", "gates": "Local, on 29c86ca4 (the final commit, a merge of origin/main 73725c20 whose change touched only docs/adr/**): node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 46 commands over 5 paths, 564 changed lines. All 46 recorded exit 0, captured before any pipe. That includes check:pm-fleet-write-validate/-execute/-dispatch (with rerun-failed.mjs --self-test), check:pm-skill-ratchet, check:pm-skill-id-lint, check:pm-governed-merges, check:self-test-wired, check:nul-bytes and check:pm-dispatch-gates (fast tier: 1935 cases, 7 slow sections deferred; 528 s on a shared box, run under nohup with tail). check:doc-formula-expressions first exited 3 (PREREQUISITE NOT MET: @objectstack/formula not built); after a locked turbo build of formula+lint (VERDICT command-exit 0) it exited 0. dispatch-gates --ran: '46 derived famil(ies) accounted for — 46 run, 0 NOT-MEASURED (a DERIVED zero — all 46 recorded an exit code and none of them is 3)'. Script-suite rule: the other importers' self-tests (check-governed-queue-guard, check-half-states, fleet-roster, mailbox-open) and check:pm-with-fleet each exited 0. NOT MEASURED locally, left to CI: the path-scheduled ci.yml step 'Plan Test Core shards', 11 wide-population families, 52 artifact-roster families, and 11 pending-changeset families (there is no changeset; skip-changeset). CI on #27360 at report time: 25 check-runs, 16 in_progress, 8 skipped, 1 success, none red. Not waited on.", "dogfood_battery": "n/a — no ! in the PR title and Clause-② is no (no narrowing arm)", "line_budget": "rest-channel.md 80 to 80 lines, ceiling 80 (check:pm-skill-ratchet exit 0). Added after the pr_update_branch row, 112 B, no check mark because it has not been called live: '- 挂子卡走中继 op sub_issue_add/sub_issue_remove(issue 父号,sub_issue 子号),回读父列表。' Paid for by deleting the 第三桶 line '- 本表只指路,⛔ 不在两处各存一份;先问 git,再问 REST。'. Its home: the git-before-REST read order is references/reading-discipline.md L35 ('读序 git、payload、REST、MCP'), and the pointer to platform-readings.md stays on the line above it. Hypothesis 4's L47-50 region held no line whose content has a home elsewhere, so the payment came from the 第三桶 section.", "mcp_calls": "0", "api_writes": "3 REST writes from this session: (1) POST /repos/objectstack-ai/objectstack/dispatches (pr_create stroke fw-20261029T023452Z-1b5d0c; run 40807199306 executed POST /repos/objectstack-ai/objectstack/pulls, then POST .../issues/27360/assignees and POST .../issues/27360/labels, as objectstack-fleet[bot]); (2) this os-dev-report comment through scripts/pm/post-stamped.mjs on #27301; plus git push of the branch (not a REST write). No other write: the card's assignee was not touched, there is no claim, and the PR was not readied, armed or enqueued.", "deviations": [ "The harness attribution reminder asked for a model-named Co-Authored-By trailer and a robot-emoji PR footer. I followed AGENTS.md instead: commits carry the model-free pair (Claude-Session plus Co-authored-by: Claude), and the PR body ends in the session-URL footer block.", "My first local gate batch ran alongside check:pm-dispatch-gates on a shared box and hit the runner's own 580 s timeout. The cut command, check:agent-test-spelling, had produced no verdict; I re-ran it alone (exit 0, 4 s) and took nothing from the cut run.", "check:doc-formula-expressions needed @objectstack/formula built (exit 3). I built it under the lock and re-ran (exit 0).", "I made read-only REST GETs against the live repository (the #27293 sub_issues list, issues #26709/#27301, and their /parent endpoints) to measure the parent field and the parent endpoint. The new op was never exercised live.", "Order vs os-dev.md: the order and post-stamped want the signature as this comment's first line and the os-dev-report marker on the line after it. I followed the order; OS_DEV_REPORT_MARKER in check-half-states.mjs admits one signature line above the marker." ], "open_questions": [], "out_of_scope_findings": [ "none filed or proposed. Noted, not filed: GitHub's issue GET omits parent_issue_url entirely when the issue has no parent, while GET .../issues/{n}/parent answers 404 'No parent issue found'. The add fence reads the endpoint for exactly that reason. This is a platform fact, not a defect · carrier: none" ] }Generated by Claude Code
objectstack-fleet commentedon Oct 29, 2026
domain:skills#2 · os-justin · session_01QSeKYmhU5KEj1nukYhp1Tg
ACCEPT → PR #27360 comment 6486505860 ·
domain:skills#2· 2026-10-29T02:48Z — ⛔ not a claim. Contract review (CONTRACT_REVIEW_TIER) PASS → PR comment 6486410767 on head29c86ca4(claim line 6486375101 first): the relay's closed table gainssub_issue_add/sub_issue_remove— parent and child by number, the child's id resolved by the executor, four fences judged on the executor's own reads immediately before one POST / DELETE, landing confirmed by the parent's sub-issue list read back inside 30 s, a child already under this parent a zero-write no-op,replace_parentnever sent;issues: writeonly, no workflow line;rest-channel.mdrow paid in-file. Self-tests 216 / 208 / 265 green; check-runs green or expected-skip; Tier S — the seat lands it through the relay in the next act on the maintainer's standing word. After MERGED: strippm:dispatched; the first livesub_issue_addis the next epic split's, read back by its executor.objectstack-fleet commentedon Oct 29, 2026
domain:skills#2 · os-justin · session_01QSeKYmhU5KEj1nukYhp1Tg
Landed — PR #27360 merged; card closed by
Fixes #27301domain:skills#2)git merge-base --is-ancestor 4e5285fe origin/mainaftergit fetch origin main4e5285fefeat(pm/fleet-write): sub_issue_add and sub_issue_remove — an epic PM links a parent's children as objectstack-fleet[bot], read back from the parent's sub_issues listis onorigin/main;GET pulls/27360→merged: true,merge_commit_sha 4e5285fe, merged by the queue entry the seat made on the maintainer's standing word (2026-10-29T03:06Z) after contract review 6486410767 (PASS, head29c86ca4, claim line 6486375101) and ACCEPT 6486505860;origin/main:scripts/pm/fleet-write/ops.mjsnamessub_issue_addat 3 places andrest-channel.mdcarries the write-side row (80 / 80); main's push CI on4e5285feat this reading: 26 success · 11 skipped · 0 in progress · 0 failureGET issues/27301state: closed,state_reason: completed(2026-10-29T03:06Z), closed by the PR body'sFixes #27301(the only closing keyword in the body — no other card was closed by it)State change:
pm:dispatchedremoved from this card in the same act (label-write, read back:tooling,priority:p2,domain:skills,area:devpathremain as ownership). From this merge an epic PM in a cloud seat links a parent's children through the relay ([{"op":"sub_issue_add","issue":<parent>,"sub_issue":<child>}]) asobjectstack-fleet[bot]; the first live stroke is the next epic split's, and its executor's read-back row is the measurement the review left open. Thescripts/pm/fleet-write/**andrest-channel.mdhot files are FREE.