Skip to content

pm-dispatch: split references/platform-readings.md by subject (step 5) #35236

Description

@objectstack-fleet

repo:objectstack#1 · os-warren · session_01JQwxbabAdcWyApSTGBYJNY

Filing gate: child of epic #35205, filed by seat 1. Ruling-ref: the maintainer said 「同意」 (2026-11-18 16:19Z) to the fleet skills review and its order of improvements (summary https://claude.ai/artifact/9TAontjAcKHUuajPPSKTGX; the three reports are in the project files under pm-orders/objectstack/skills-review-2026-11-18/). ⛔ Not a claim.

The order is one step, one PR, starting after the rule PR for #35207 merges: skill ratchet by bytes → #35215 (clear current values) → contradictions → slim entry files → lanes become seats plus epics (text → scripts → lane files) → split platform-readings → prose into scripts → each repo's rules back to its own repo, cloud skills and dogfood.

What to do

Step 5. Split .claude/skills/pm-dispatch/references/platform-readings.md into smaller files by subject, so a run reads only the readings it needs. Measured readings keep their content. Readings that no longer describe the platform are listed in the PR for the maintainer, not silently dropped.

Done when

No single readings file carries unrelated subjects, every pointer to the old file resolves, and the skill gates pass.

Activity

  1. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    epic:#35205#1 · os-warren · session_01JQwxbYJqoPAoSpWTMpH1yq

    State: pm:blocked → pm:queue (the chain is cut) · 2026-11-19

    Was: pm:blocked on #35235. Splitting references/platform-readings.md shares no file with the lane reorganisation.

    Order: Warren, 2026-11-19 00:32Z 「可以拆的都同意。」 (item 6: the fleet skills clean-up chain becomes parallel; #35229 starts now). Rule: #35259 item 6 — fleet cards block each other only when they edit the same files. The new order is on #35205.


    Generated by Claude Code

  2. added
    pm:queueReady for the PM dispatch loop
    and removed
    pm:blockedBlocked by another issue/PR — body carries Blocked-by: #N
    on Nov 19, 2026
  3. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#1 · os-warren · session_01JQwxbYHmpLgwQvpwmbGvLL

    Grade: p2 → p1 — the maintainer's manual grade (no triage or seat moves it).

    Source: Warren, 2026-11-19 in the objectstack cloud os project thread: 04:51Z 「哪几个没做的,改到 p1」 — the not-yet-started fleet skill cards #35259, #35298, #35215, #35229–#35232 and #35234–#35238; 04:52Z 「不是让你做」 — relabel only. State label, assignee and Blocked-by lines are unchanged; nobody claims the card by this comment.

  4. added
    priority:p1High: required for production / M2
    and removed on Nov 19, 2026
  5. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#1 · os-warren · session_01JQwxbYHmpLgwQvpwmbGvLL

    Claim: PM loop round 1
    Session: session_01JQwxbYHmpLgwQvpwmbGvLL
    Thread session: session_01JQwxbZj3NQLFYjgQqTUEtj
    Account: os-warren
    Branch: claude/project-thread-ga58bl
    Worktree: objectstack-issue-35236
    Domain: domain:skills
    Seat: repo:objectstack#1
    File surface: .claude/skills/pm-dispatch/references/platform-readings.md → new files beside it, plus pointer lines to it (stop on breach; explain in the report)
    Container & model: M, mode:thread, model: opus — dispatch-gates --tier: no path-derived mandate; the one SKILL.md pointer line is a one-line mechanical exit, effort: high
    Clause-②: no
    Metadata-protocol: none
    Responsibility: n/a — not a defect card
    Thread-read: 7000195656
    Premise: holds
    Serial constraints cleared: card thread for #35259/#35298/#35230 (branch claude/project-thread-rothb5) edits SKILL.md and other references at the same time; pointer-line conflicts rebase on whichever lands first.

    Folded with #35215 on this branch (#35215 posts a list on its card, no file change); the PR carries Fixes #35236 only.

    Source of the dispatch: Warren, 2026-11-19 in the objectstack cloud os project thread: 04:54Z 「那你还是直接派发吧」; 04:55Z 「为什么不能合并派发」.

  6. added
    pm:dispatchedDispatched to a dev agent by /pm-dispatch
    and removed
    pm:queueReady for the PM dispatch loop
    on Nov 19, 2026
  7. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#1 · os-warren · session_01JQwxbYHmpLgwQvpwmbGvLL

    Release: dispatch stopped before a PR; nothing landed. Any work so far stays on branch claude/project-thread-ga58bl for whoever picks the card up. Back to pm:queue at p1, unassigned.

    Source: Warren, 2026-11-19 05:26Z in the objectstack cloud os project chat: 「你没有token了,没让你派这么多」 — dispatch stops until he says so; the card thread for #35259/#35298/#35230 is the only one that continues.

  8. added
    pm:queueReady for the PM dispatch loop
    and removed
    pm:dispatchedDispatched to a dev agent by /pm-dispatch
    on Nov 19, 2026
  9. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#5 · os-bill · session_01Utp6CaAqjP52wVTzs99H5G

    Claim: PM loop round 1 · area fleet-1 (lease:fleet-1; the maintainer, verbatim, Bill 2026-11-19 05:32Z 「这些你都接手」 for seat 1's hand-off, mailbox #35216 comment 7000702066) · 2026-11-19T05:44Z
    Session: session_01Utp6CaAqjP52wVTzs99H5G
    Account: os-bill
    Branch: claude/project-thread-ga58bl
    Worktree: objectstack-issue-35236
    Domain: domain:skills
    Seat: repo:objectstack#5
    File surface: .claude/skills/pm-dispatch/references/platform-readings.md → the nine readings-*.md files beside it, plus the pointer lines to it (SKILL.md, os-dev.md, AGENTS.md, references), scripts/check-required-contexts.mjs paths, the ratchet rows for the new files (stop on breach; explain in the report)
    Container & model: M, mode:subagent, model: default — dispatch-gates --tier: no path-derived mandate (the SKILL.md / os-dev.md / AGENTS.md edits are one-line pointer swaps, a mechanical exit), effort: high
    Clause-②: no
    Metadata-protocol: none
    Responsibility: n/a — not a defect card
    Thread-read: 7000665868
    Premise: holds
    Resume-from: claude/project-thread-ga58bl@0642fef5f3
    Serial constraints cleared: draft PR #35337 already carries the work (head 0642fef5f3, mergeable clean with main). Same area: G1 (#35259/#35298/#35230, claude/project-thread-rothb5) touches the same SKILL.md / os-dev.md files, overlap check clean; #35229 (claude/project-thread-xtbyqx) rewrites scripts/pm/check-skill-line-ratchet.mjs and conflicts with this PR's ratchet rows — #35229 queues behind this PR and re-measures.
    Pre-claim lane: area:fleet / domain:skills — .claude/skills/pm-dispatch/references/**
    Pre-claim deps: no Blocked-by:; the step order on the card is cut to parallel by the same-file rule (#35230 comment 6996492814)
    Prior rulings read: the maintainer's 「同意」 to the skills review order (quoted on the card); docs/adr/** grep platform-readings → none on point; no conflicting record

    Readings the PR lists as possibly stale go to the maintainer with #35215's keep/delete list (the same question: text that no longer holds), not decided in this PR.

  10. added
    pm:dispatchedDispatched to a dev agent by /pm-dispatch
    and removed
    pm:queueReady for the PM dispatch loop
    on Nov 19, 2026
  11. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#5 · os-bill · session_01Utp6CaAqjP52wVTzs99H5G
    os-dev-report

    {
      "issue": 35236,
      "status": "done",
      "branch": "claude/project-thread-ga58bl",
      "pr": "https://github.com/objectstack-ai/objectstack/pull/35337",
      "session": "session_01Utp6CaAqjP52wVTzs99H5G",
      "premise_still_valid": true,
      "head_sha": "803f13aafdf9414b77fb993efad8d6b7cf55e6bf",
      "summary": "Resumed from claude/project-thread-ga58bl@0642fef5f3 and did not redo the split. I checked it: every reading moved verbatim, and every pointer resolves except one, which I fixed in commit 803f13aafd. rest-channel.md:64 said 'see the quota section' with no file name, and that section is now readings-channels.md; the line now names that file at 119 of 120 bytes, with the line ceiling unchanged. The PR body was rewritten through the relay (issue_patch, read back identical, 11666 bytes): the attribution head lines, Fixes, Clause-2 and Metadata-protocol lines are kept; the maintainer brief is now four business lines ending in 无; Look first, Two-minute check and the stale-readings list moved into the English part. The PR stays draft, with no reviewers and no queue.",
      "tests": "Verbatim proof: sorted multiset of non-blank lines, old file at merge base 81c094779 (unchanged on origin/main) vs the nine readings-*.md plus the index. 447 old lines and 463 new. Only in old: the title and the 3 section headings. Only in new: 9 file titles, the index title, the index lead line and 9 rows. Every other line is byte-identical. Item-level (each bullet with its continuation lines): 439 old items, 0 missing, 0 altered, and the 10 extra are the index's own lines. Pointer check: git grep platform-readings leaves only the reading-discipline.md index pointer, historical records and fixtures (check-governed-merges.mjs:7001, check-governed-queue-guard.mjs:3696/:5436), ratchet ceiling rows and verbatim ruling quotations. I grepped for the old section names (配额段, 队列段, 读数陷阱, 队列成员资格与 auto-merge): after the fix, the only hits are the verbatim ruling quote at check-skill-line-ratchet.mjs:1233 and the comment at :1219, which I left alone because it is outside the fence and reaches the file through the index. I read each swapped pointer against the content of its target file and found no mismatch.",
      "gates": "Local, head 803f13aafd. node scripts/pm/dispatch-gates.mjs --commands exit 0 (80 commands) plus check-published-list-mirrors (run and --self-test) and check-nul-bytes = 82 commands; 79 exit 0. Named gates: node scripts/pm/check-skill-line-ratchet.mjs exit 0; pnpm check:pm-skill-ratchet exit 0; pnpm check:pm-skill-id-lint exit 0 (23 self-test cases, 129 files clean); pnpm check:required-contexts (self-test 190 assertions, then the run: 11 contexts, 5 surfaces) exit 0; node scripts/check-published-list-mirrors.mjs exit 0, --self-test exit 0 (37 cases). Gates for scripts that name the old file, all exit 0: check:pm-half-states, check:pm-post-stamped, check:pm-label-write, check:pm-fleet-write-dispatch (includes rerun-failed --self-test), check:pm-fleet-write-execute, check:pm-governed-merges, check-governed-queue-guard --self-test. pnpm check:pm-dispatch-gates ran detached: fast tier 2249 cases pass, 494.4s, exit 0. NOT MEASURED, exit 3 PREREQUISITE NOT MET because nothing is built in this worktree: check:doc-formula-expressions and check:i18n-coverage. Exit 1 not from this diff: check:platform-checklist (seed-mode-matrix VARIANTS STALE, SeedMode has 0 members, which comes from reading an unbuilt packages/spec). CI's 'Platform checklist gate sweep' and 'Lint & Repo Gates' passed on this head. --ran reconciliation was not run separately: the run list is exactly the derived list plus 3. CI check-runs on 803f13aafd: success for Lint & Repo Gates, TypeScript Type Check, Governed Surface Queue Guard, Test UI, Dogfood Regression Gate, Platform checklist gate sweep, PR-body guards and Live required-set sweep; in_progress for Test Core (1/10) and Live half-state sweep; Build Core, Lint UI, Type Check UI, Build & E2E UI and Live E2E UI were skipped by the selector. On 0642fef5f3 every required context was success or skipped.",
      "overlap_verdict": "node scripts/pm/overlap-check.mjs --branch=claude/project-thread-ga58bl --exclude-pr=35337, VERDICT overlap exit 1: it merges cleanly with origin/main e4bfa339ab and with #35335, which shares check-skill-line-ratchet.mjs and label-write.mjs. #34110 (os-steve, non-draft) conflicts on .claude/skills/pm-dispatch/references/triage-duties.md, which that PR also edits; that conflict is not mine to fix. #35229 has no open PR head in the overlap set; per the dispatch it queues behind this PR on the ratchet file. GitHub reports mergeable true, so I did not merge main (token rule).",
      "files_changed": "29 vs merge base: os-dev.md, SKILL.md, AGENTS.md, references/{platform-readings, readings-channels, readings-ci, readings-issue-fields, readings-page, readings-queue, readings-search, readings-session, readings-shell, readings-write-shapes, area-lease, landing-operations, reading-discipline, rest-channel, triage-duties, lanes/director, lanes/server}.md, scripts/check-required-contexts.mjs, scripts/pm/{check-governed-queue-guard, check-half-states, check-skill-line-ratchet, label-write, post-stamped, rerun-failed, fleet-write/dispatch, fleet-write/ops}.mjs. This round added only references/rest-channel.md:64.",
      "stale_readings": [
        "readings-shell.md:14 + readings-channels.md:47: they say the container has no gh, but command -v gh answers /usr/local/bin/gh in this cloud container",
        "readings-queue.md:43-44 + readings-channels.md:63: readings of the ccr undraft/arming route, which line 43 itself calls retired (已退役); undraft and arming now go through relay ops",
        "readings-queue.md:40 + readings-write-shapes.md:42: the MCP update_pull_request channel, which the lines themselves say lock 1 refuses (读作历史)",
        "readings-channels.md:34-38,40: claude[bot] vs user token classes and the dated account flips; fleet writes now land as the fleet bot through the relay",
        "readings-write-shapes.md:65-66: footers under MCP issue_write create/update; seats and devs no longer write through MCP",
        "readings-channels.md:29: REST rerun-failed-jobs 403 vs MCP 201; re-runs now go through the relay's run_rerun_failed (readings-ci.md:5)",
        "readings-session.md:37-38: Routine sessions and the connectors parameter; line 39 says it is unmeasured (未实测)"
      ],
      "dogfood_battery": "n/a — no ! in the PR title and no (narrowing) arm",
      "mcp_calls": "0",
      "api_writes": "2, both through the relay (fleet-write dispatch, POST /repos/objectstack-ai/objectstack/dispatches): (1) issue_patch PATCH /repos/objectstack-ai/objectstack/issues/35337 body, run 44469276659 success, read back identical; (2) the os-dev-report comment POST /repos/objectstack-ai/objectstack/issues/35236/comments via post-stamped.mjs. Plus one git push (0642fef5f3..803f13aafd), which is not a REST write.",
      "deviations": [
        "Commit trailers: the dispatch asked for a Co-Authored-By trailer that names a model. AGENTS.md and os-dev.md require the model-free pair, and the pre-push hook refuses a model id, so commit 803f13aafd carries Claude-Session plus the model-free Co-authored-by trailer (name Claude, the noreply address).",
        "PR body ending: the dispatch said to end with the Generated-with line, a blank line and the session URL. The relay's issue_patch door refuses a PR-body rewrite unless its last non-empty line is the PR_BODY_TRAILER_SHAPE signature comment. The body therefore ends with those lines followed by that signed trailer, spelled 'signed: repo:objectstack#5 · os-bill · session_01Utp6CaAqjP52wVTzs99H5G' inside an HTML comment (os-dev.md allows that trailer when an order calls for it).",
        "Fence: the previous thread's commit 0642fef5f3 also edits pointer comments and messages in scripts/pm/{check-governed-queue-guard, check-half-states, label-write, post-stamped, rerun-failed, fleet-write/dispatch, fleet-write/ops}.mjs. The dispatch fence names only scripts/check-required-contexts.mjs and the ratchet rows among scripts. These are pure pointer swaps, which the card's done-when ('every pointer resolves') needs, so I kept them and did not stop; this is flagged here for the seat."
      ],
      "open_questions": [],
      "not_done": [
        "check:doc-formula-expressions and check:i18n-coverage were not measured locally (unbuilt prerequisites; CI builds them). check:platform-checklist is red locally only, from the unbuilt spec, and green in CI.",
        "The comment at check-skill-line-ratchet.mjs:1219 still says 'the policy lives in platform-readings.md's quota section'. I left it because it is outside the fence, #35229 rewrites that file, and the comment reaches the policy through the index.",
        "Test Core (1/10) and Live half-state sweep were still in_progress at report time."
      ],
      "security_findings": "none",
      "out_of_scope_findings": [
        "carrier: none · noted, not filed: the overlap-check found #34110 conflicting with this PR on triage-duties.md. Whichever lands second re-applies its line; this is not a defect."
      ]
    }
  12. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#5 · os-bill · session_01Utp6CbUb7TcZK1XUAddUmQ

    ACCEPT → PR #35337 comment 7001853325 · head 93f0ef1d · 2026-11-19T06:42Z. Contract review delta PASS 7001841085 (same head). Tier H (AGENTS.md): awaiting a maintainer merge or approval (reviewers os-zhuang, hotlong); close-out here on MERGED.

  13. objectstack-fleet commented on Nov 19, 2026

    @objectstack-fleet
    ContributorAuthor

    repo:objectstack#5 · os-bill · session_01Utp6CbUb7TcZK1XUAddUmQ

    Close-out · PR #35337 MERGED 2013c3828a · 2026-11-19T08:06Z. Delivered: references/platform-readings.md split into per-subject readings-*.md files with the original as their index; pointers updated across the skill, rest-channel.md and AGENTS.md. Carried elsewhere: the stale ratchet credit re-measures in #35229 (7001508654); the dated-reading cleanup is #35215 G-D, its own PR. pm:dispatched off.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions