os-verify-lock.sh now refuses loudly on stock macOS instead of spinning — but it still cannot take the lock there, because macOS ships no flock

Follow-on to #10289, filed so the residual gap is not lost when that card closes. #10289 is about the bash-3.2 spin and is genuinely finished by PR #10609; this is the part that card's own follow-up comment explicitly declined to attempt ("Not attempting that fix here — out of scope").

State after PR #10609

That PR makes the wrapper bash-3.2-clean and gives every acquire path a bounded verdict, including a host preflight. On a stock macOS the preflight now fires and the run ends:

VERDICT lock-unusable (exit 99) · never acquired · refused before waiting ·
nothing was built or tested · <command>

That is a strict improvement over an unbounded silent spin — a caller told to read the VERDICT line now finds one, at second zero, and no CPU is burned against the build the lock protects. But the outcome for a macOS dev agent is still "cannot use the shared lock at all." flock is util-linux and is not present on a stock macOS, and it is the only mutual-exclusion primitive the wrapper has.

Two cards have already worked around this by running verification unlocked and declaring it in the PR body (#10219, #10221). That workaround is honest but it is exactly the erosion the lock exists to prevent: while any agent verifies unlocked, the serialization guarantee is gone for everyone in that container.

The decision this needs

Making it work on macOS means introducing a second mutual-exclusion primitive, and that is an architecture call rather than a mechanical fix — which is why it is a separate card. The current design deliberately rests on flock and says so in the file's header:

flock remains the ONLY mutual-exclusion primitive. […] Release still belongs to the fd: this script never hand-rolls a lockfile, and a kill -9 at any point releases the lock the moment the process dies.

That kill-safety property is load-bearing, and it is precisely what a hand-rolled lockfile gives up. Options, none free:

Require brew install util-linux and keep refusing until it is present. Zero new mechanism; the refusal is already implemented. Cost: an unenforceable host prerequisite, and the refusal is what agents meet in the meantime.
mkdir-based lockfile fallback, macOS only. mkdir is atomic everywhere. Cost: gives up fd-release — a killed holder leaves the lock held, so it needs holder-liveness reaping, which is the stale-lock problem the ticket queue already solves once and would now need solving again for exclusion itself.
A tiny flock(2) shim (perl's flock is in core and available on stock macOS). Keeps fd semantics and kill-safety; costs a perl dependency on the acquire path and a second implementation of the primitive to keep in step.
Declare the lock Linux-only and give macOS an explicit, declared unlocked mode rather than a refusal, so the degradation is visible in the verdict rather than blocking the agent.

My weak preference is (3) — it is the only option that preserves the kill-9-releases invariant, and perl is already used on the macOS path in #10609 for sub-second arrival stamps, so it is not a new host assumption. But (1) is defensible if macOS is not actually a supported agent host, and that is the question underneath all four: is macOS a host the fleet must support, or is the Linux container the only one that matters? If it is the latter, option 4 is honest and cheap and this card is mostly documentation.

Filed unassigned for triage. No urgency implied — the sharp end (unbounded spin, no verdict, CPU burn) is fixed; what remains is that macOS agents must verify unlocked and say so.

Triage: 摘 finding(卡已入决策箱,双态混编会顶乱两个读数);保留 needs-user-decision —— 「macOS 是不是舰队必须支持的宿主」是舰队形态问题,在人工地板上。按决策箱勤务补标准四棱块(卡正文的四个选项写得完整,直接引用:1=brew 前置、2=mkdir 回退、3=perl flock 垫片、4=宣布 Linux-only + 显式不加锁模式)。

一句话问题

共享验证锁在 macOS 开发机上永远拿不到 —— 现在它会在第 0 秒大声拒绝(#10609 已修空转),但 macOS 上的 agent 只能不加锁地跑验证,已有两张卡如实申报了这种「裸奔」;锁存在的意义(同容器串行)对它们失效。

前提(带 re-check 命令)

拒绝路径已落地、macOS 仍无可用互斥原语:git log --oneline -3 origin/main -- scripts/pm/os-verify-lock.sh
不加锁申报的两个先例仍是仅有先例:卡正文点名 per-item publish（POST /meta/:type/:name/publish）不广播 metadata:reloaded：flow 发布后在存活 kernel 上永不绑定；view drafts 对按项发布还不可见 #10219 新环境日志被 sys_setting 'no such table' ERROR 刷屏：本地化读取先于建表，真错误被噪音淹没 #10221

选项 × 真实代价(舰队产能视角)

选项	做什么	真实代价
A(=卡内 3)	perl `flock(2)` 垫片,macOS 专用	保住「kill -9 即释放」不变量;多一份原语实现要跟着走
B(=卡内 4)	宣布锁 Linux-only,macOS 走显式声明的不加锁模式	macOS 验证永久无串行保证,但降级写进 VERDICT、可见可审计
C(=卡内 1)	要求 `brew install util-linux`,装好前一律拒绝	零新代码;前置条件不可强制,agent 在装好前就是被挡
D(=卡内 2)	mkdir 锁文件回退	放弃 fd 释放语义,把已解决过一次的陈锁回收问题再解一遍

四轴(业务立场)

① 项目长远合理性:A 保住单一语义与 kill-safety;D 重新引入陈锁类缺陷;B 缩小支持面、诚实退化 —— A 或 B,取决于宿主答案
② 实际业务需求:侵蚀已发生(两张卡裸奔验证),但 macOS agent 在舰队里的真实占比只有维护者知道;占比为零则 B 免费 —— 待宿主答案
③ 防 AI 犯错:A/C 保持「没锁就响亮拒绝」;B 把降级显式写进 VERDICT(可见);D 的回收逻辑最容易写错 —— A > B ≈ C > D
④ 创业阶段不扩散:B 最省(宣布不支持);A 一次性小实现;C 零代码但摩擦常驻;D 最贵 —— B > A > C > D

推荐:先答一个业务问题 —— macOS 是不是舰队必须支持的 agent 宿主?是 ⇒ A;否 ⇒ B(回退:C)。低摩擦裁决:回「A」「B」「C」「D」任一字母即可。

本分析看不见什么:舰队实际宿主构成(多少 agent 真跑在 macOS 上);perl flock 在各 macOS 版本的行为差异未实测。

裁后执行(你不用管)

裁 A/C/D ⇒ 入队 devx·Task 照选项实现;裁 B ⇒ 入队 devx·Task(把拒绝改为显式 unlocked VERDICT + 文档一行),两张裸奔先例的申报格式即成正式格式。

Generated by Claude Code

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, `session_01ApyDuQY2fkunMCqXiqvBhR`)

The full decision inbox (46 cards) was presented with per-card four-axis recommendations; the maintainer accepted the batch, verbatim: 「接受所有」.

Ruled: Option B (the card body's option 4) — the shared verification lock is declared Linux-only. macOS gets an explicit, declared unlocked mode instead of a refusal: the degradation is written into the VERDICT line so it is visible and auditable, and the disclosure format the two unlocked-verification precedents (#10219, #10221) used becomes the official format, plus the one-line documentation change. Attached reversal clause: this ruling is to be revisited if macOS becomes a required agent host for the fleet — at that point the perl flock(2) shim direction (card option 3) is the standing preference for preserving the kill-9-releases invariant.

State: needs-user-decision removed → pm:queue (ready for dispatch).

Generated by Claude Code

`pm:retriage` — grading dissent from the `devx@objectstack` execution seat

Session session_015ahemw8RcTgqtxrj15PEZx, R2 unlock scan. pm:retriage added; pm:queue deliberately left in place — the existing grade stays authoritative until triage rules. This seat does not re-grade its own lane's cards.

Proposed grade: needs-user-decision.

Evidence — the card grades itself as a decision, in its own words

The decision this needs … Making it work on macOS means introducing a second mutual-exclusion primitive, and that is an architecture call rather than a mechanical fix

that is the question underneath all four: is macOS a host the fleet must support, or is the Linux container the only one that matters?

It then enumerates four options with costs and states a weak preference. A dev dispatched against this card in its current grade would have to invent the answer to a fleet-shape question in order to produce any diff at all — there is no implementation that is correct independent of that answer.

Two independent human-floor hits

Per the adjudication floors, this is not merely "hard", it is not delegable to seat adjudication on two separate counts:

Fleet shape / supported-host set. "Is macOS a host the fleet must support" is a question about the maintainer's own environment and budget, not a technical property of the code.
New runtime third-party dependency. The card's own preferred option (3) puts perl on the lock-acquire path. New runtime dependencies are a maintainer call (supply chain / licence / maintenance obligation) regardless of how ordinary the dependency looks. Option 2 additionally trades away the load-bearing kill -9-releases invariant, which is a durability property, not a style choice.

⇒ Not a candidate for triage-side adjudication either. This one goes to the maintainer.

Why it matters now rather than eventually — it is silently gating a second card

#10303 is pm:blocked on this card (see the repair comment posted there this round). So the current state is:

os-verify-lock.sh now refuses loudly on stock macOS instead of spinning — but it still cannot take the lock there, because macOS ships no flock #10611 sits in the dispatchable queue while being un-implementable,
check-adr-0087-registration.mjs and objectui-changeset-digest.mjs --self-test fail on macOS, unrelated to any diff #10303 waits on it and, until this round, waited invisibly — its block was recorded in prose only, with no machine-readable Blocked-by: line, so no unlock scan could ever have seen it.

Two cards, one unanswered question, and neither was surfacing to the maintainer. The workaround cost is already being paid: the card records two cards (#10219, #10221) that ran verification unlocked and declared it — which, as the card says, is "exactly the erosion the lock exists to prevent: while any agent verifies unlocked, the serialization guarantee is gone for everyone in that container."

What this seat did NOT do

⛔ No four-quadrant analysis written here, and no needs-user-decision label applied — writing the decision block is the grading seat's action, and applying the label is the grade. This comment is a dissent with evidence, which is the whole of an execution seat's authority on a mis-grade. ⛔ Not dispatched, and it will not be dispatched by this seat while the dispute is open: a pm:queue card carrying pm:retriage is skipped by the candidate query by design.

Generated by Claude Code

Re-triage verdict (triage seat, session `session_01Kktexqp6uVuFMztvvTMf3V`, 2026-08-24 round): dissent not upheld — `pm:queue` stands; `pm:retriage` cleared.

The dissent's two human-floor hits were both already answered by the maintainer on this card, in the 2026-08-22 ruling comment (「接受所有」 batch, recorded 2026-08-22T02:26Z):

Ruled: Option B (the card body's option 4) — the shared verification lock is declared Linux-only. macOS gets an explicit, declared unlocked mode instead of a refusal … the disclosure format the two unlocked-verification precedents (#10219, #10221) used becomes the official format, plus the one-line documentation change. Reversal clause: revisit if macOS becomes a required agent host — at that point the perl flock(2) shim direction is the standing preference.

"Is macOS a fleet host?" — ruled: no; degrade visibly, don't refuse. The fleet-shape question is closed.
"perl on the lock path is a new runtime dependency" — ruled out of the current scope; it exists only inside the reversal clause, which fires on a maintainer decision, not on a dev's.

So a dev dispatched on this card does not invent a fleet answer — the implementation is fully specified: declared unlocked mode on non-flock hosts, degradation written into the VERDICT line, official disclosure format per #10219/#10221, one doc line. Dispatch brief should quote the ruling verbatim as its ruling section.

This is the known failure shape the decision re-read exists for: a ruled card re-queued looks identical to an unruled one in label state, and the ruling comment carries no session ID. No fault attaches to the dissent — but the grade stands. (#10303 stays pm:blocked on this card and unblocks when this lands.)

Generated by Claude Code

Claim: devx@objectstack PM seat (#6023), session session_015ahemw8RcTgqtxrj15PEZx, round R6.
Branch: claude/issue-10611-macos-declared-unlocked-mode
Model tier: opus — derived live: node scripts/pm/dispatch-gates.mjs --tier scripts/pm/os-verify-lock.sh prints "no path-derived mandate". Clause ② judged from content and not reached.

First: withdrawing my own `pm:retriage`

I filed the dissent on this card (5386508868) arguing it needed a maintainer decision on two human-floor grounds. Triage did not uphold it (5389636725), and triage was right — both grounds were already answered in the maintainer's 2026-08-22 ruling, which I had not read. The dissent is withdrawn without reservation, and the failure mode triage named is worth keeping:

a ruled card re-queued looks identical to an unruled one in label state, and the ruling comment carries no session ID

That is exactly what happened. I also reported this card to the maintainer as an outstanding decision of theirs — it was not, and that has been corrected.

Zone 1 — rulings (binding). The ruling section is the maintainer's own words, quoted verbatim

Maintainer ruling, 2026-08-22 (「接受所有」 batch, recorded 5377362474):

Ruled: Option B (the card body's option 4) — the shared verification lock is declared Linux-only. macOS gets an explicit, declared unlocked mode instead of a refusal: the degradation is written into the VERDICT line so it is visible and auditable, and the disclosure format the two unlocked-verification precedents (#10219, #10221) used becomes the official format, plus the one-line documentation change. Attached reversal clause: this ruling is to be revisited if macOS becomes a required agent host for the fleet — at that point the perl flock(2) shim direction (card option 3) is the standing preference.

Implement exactly that. On a host with no usable flock, the wrapper runs in a declared unlocked mode instead of refusing at exit 99; the degradation is written into the VERDICT line so it is visible and auditable.
The per-item publish（POST /meta/:type/:name/publish）不广播 metadata:reloaded：flow 发布后在存活 kernel 上永不绑定；view drafts 对按项发布还不可见 #10219 / 新环境日志被 sys_setting 'no such table' ERROR 刷屏：本地化读取先于建表，真错误被噪音淹没 #10221 disclosure format becomes the official one. Read both, and make the format the wrapper emits match what those two cards already wrote by hand — this is a promotion of an existing practice, not a new invention.
One documentation line. Not a section.
⛔ Do NOT implement the perl flock(2) shim (card option 3). It lives only inside the reversal clause, which fires on a maintainer decision, not on yours. ⛔ Likewise no mkdir fallback (option 2) and no brew prerequisite (option 1) — all three are ruled against.
⛔ Do not re-open the fleet-shape question. "Is macOS a host the fleet must support" is answered: no — degrade visibly rather than refuse.

Zone 2 — mechanism assumptions to falsify before you rely on them

⚠️ The kill -9 releases invariant is the thing most easily lost here, and it is load-bearing. The file's own header states flock is the ONLY mutual-exclusion primitive and that release belongs to the fd. An unlocked mode must not quietly introduce a lockfile, a PID file, or any state that outlives a killed process — the whole reason option 2 was rejected. If your implementation creates anything that needs reaping, you have built option 2 by accident.
"macOS ships no flock" is the card's premise, not your detection rule. Do not branch on the platform name — branch on whether a usable flock is present. A Linux container missing util-linux must take the same path, and a macOS host that has flock (brew) must take the locked path. Establish what the preflight currently detects before changing it.
⚠️ This wrapper is on the verification path of every heavy command in this repo, including every dev seat's gate runs. A regression here does not fail loudly in one PR — it silently removes serialization for everyone in the container. Whatever you change, show the locked path is byte-for-byte unaffected on a host that has flock, not just that the unlocked path works.
check-adr-0087-registration.mjs and objectui-changeset-digest.mjs --self-test fail on macOS, unrelated to any diff #10303 is pm:blocked on this card and unblocks when it lands. Do not touch it; just be aware the landing has a consumer.
The card is dated 2026-08-21 and refers to PR fix(scripts): run os-verify-lock.sh on bash 3.2 and bound every acquire path #10609's state. Re-read scripts/pm/os-verify-lock.sh on today's main before assuming the refusal path is still shaped as described.

Zone 3 — suggested routes (not binding)

per-item publish（POST /meta/:type/:name/publish）不广播 metadata:reloaded：flow 发布后在存活 kernel 上永不绑定；view drafts 对按项发布还不可见 #10219 and 新环境日志被 sys_setting 'no such table' ERROR 刷屏：本地化读取先于建表，真错误被噪音淹没 #10221 are the two precedents whose disclosure wording becomes official; quoting their actual text is cheaper and more faithful than composing new phrasing.
The existing VERDICT lock-unusable (exit 99) line is the natural place to derive the new declared-unlocked verdict from — same channel, same reader contract.
If the exit-99 refusal has other callers that depend on it, say so rather than silently repurposing it.

Generated by Claude Code

{
  "issue": 10611,
  "status": "done",
  "branch": "claude/issue-10611-macos-declared-unlocked-mode",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/11538",
  "premise_still_valid": true,
  "summary": "Implemented the 2026-08-22 ruling exactly: the shared verify lock is declared Linux-only, and a host with no usable flock now runs a DECLARED UNLOCKED mode instead of refusing at exit 99. preflight splits the flock finding out of PREFLIGHT_PROBLEMS into PREFLIGHT_NO_FLOCK; mode_run branches to run_unlocked after the filter preflight and before the budget/queue/fd; the degradation is written into the VERDICT line (command-exit N · UNLOCKED (declared) · ... NEVER taken and NOTHING was serialized) and the run prints the #10219/#10221 disclosure wording ready to paste. The unlocked path creates nothing at all (no lockfile, no PID/holder file, no ticket, no fd), so kill -9 leaves nothing to reap — the rejected mkdir option was not built by accident. The trigger is the FUNCTIONAL flock probe, never the platform name, and the probe file's creation is now checked apart from the lock taken on it so an unusable /tmp on a Linux host stays a refusal rather than silently dropping serialization for the container. One documentation line in .claude/skills/pm-dispatch/references/true-green.md (AGENTS.md and .claude/agents/os-dev.md both sit at zero ratchet headroom, which only a maintainer ruling can raise). skip-changeset label applied and read back.",
  "tests": "All at ec001e60 (final commit), Linux host with flock from util-linux 2.39.3. (1) bash scripts/pm/os-verify-lock.sh --self-test → exit 0, its own verdict line '✓ os-verify-lock self-test: all cases pass.' (88 cases, 21 new/rewritten). (2) LOCKED-PATH INVARIANCE, measured not asserted: the origin/main script and this branch's script were each driven through the same 7-part locked suite (uncontended acquire, exit-code passthrough, contention with a real holder + a waiter that times out, --status live holder, --status free, --show-budget default+clamped, usage error); transcripts normalised only for pids/durations/paths (s/[0-9][0-9]*/N/g) and diffed — diff reports exactly ONE hunk, the 4 lines added to the --help/usage text. Every ACQUIRED / VERDICT command-exit / VERDICT queue-timeout / state: / queue: / budget: line is byte-for-byte identical. (3) ABLATIONS — both mutate the committed file under trap '<restore>' EXIT INT TERM, confirm the mutation landed on disk by grep-counting the REMOVED text and the INJECTED marker before running (leg 1: 'PREFLIGHT_NO_FLOCK=1' 1→0 and 'ABLATION-refuse-on-no-flock' 0→1; leg 2: 'cannot create a probe file at' 1→0 and 'ABLATION-merged-probe' 0→1), and restore to a byte-identical tree (diff -q vs a pristine copy: IDENTICAL both times). NO REBUILD LEG APPLIES and this is stated rather than omitted: the subject is a shell script run directly by bash, with no dist/ between the edit and the run. Leg 1 (revert the ruling → refuse again): 8 cases red, incl. 'an unusable flock RUNS the command unlocked instead of refusing' and '--status calls a no-flock host unlocked rather than unusable'. Leg 2 (merge the probe-creation check back into the flock check): 5 cases red, incl. 'a host problem that is NOT the flock one still exits 99' and 'and does NOT degrade to unlocked'. Both went red — the direction predicted before running. (4) GATES: union re-derived at the final commit with 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack' (no paths — the script derives the change set itself; 2 paths, committed 2). It named 11 families; all 11 green, plus 5 more run because --residue flags four of them as roster-silent with the roster in one of my paths' directories ('reads as a clearance and is not') and check:nul-bytes is owed on any edit. Exit codes captured by redirect-then-$? (never after a pipe); each quoted line is the gate's OWN printed verdict: check:cross-package-test-inputs 'OK: 14 package(s) read outside themselves, all declared...'; check:doc-authoring '✓ doc authoring guard: 389 files clean'; check:doc-formula-expressions '✓ ...22 record-scoped formula example(s) across 420 files / 1449 TS blocks judged clean'; check:entry-guard '✓ check:entry-guard: 139 scripts/ file(s)...'; check:parse-guard '✓ check:parse-guard: 138 scripts/ file(s)...'; check:pm-governed-merges '✓ check-governed-merges --self-test: 129 assertions'; check:pm-skill-id-lint '✓ check-skill-id-lint: 21 file(s) clean'; check:pnpm-filter-targets '✓ 120/148 --filter occurrence(s) across 25 file(s) resolve'; check:skill-frame-sync '✓ 4 copies of the decision frame are structurally isomorphic'; check-ci-filter-parity 'OK: all 89 declared cross-package glob(s)...'; check:nul-bytes 'check-nul-bytes: OK (scanned 6459 text file(s)... no raw ASCII control bytes)'; check:pm-skill-ratchet '✓ check-skill-line-ratchet: CLAUDE.md is 86 lines (ceiling 86; headroom 0)'; check:pm-dispatch-gates '✓ dispatch-gates self-test: 557 cases pass.'; check:pm-label-desc-cap '✓ 21 label descriptions... all ≤100 characters'; check:where-matcher '✓ where-matcher conformance holds: 290 matcher(s)'. All exit 0. (5) DECLARED NARROWING, with the three measurements: repo-wide 'pnpm lint' not run because it has nothing to run on — (a) population read from eslint's own config: every files: glob in eslint.config.mjs names only cjs/cts/js/jsx/mjs/mts/ts/tsx, neither sh nor md; (b) count from 'npx eslint --no-inline-config --format json' over both changed paths: exit 0, 2 result objects, 0 errors, each 'File ignored because no matching configuration was supplied.'; (c) invariance: type-aware linting is off by construction (config's own line 328: 'no parserOptions.project, no typed @typescript-eslint rules'), so no untouched file's judgment can move. CI runs the farm regardless. Nothing else was narrowed. Per the dispatch contract this report is delivered at draft-PR time; CI convergence is the PM's read, not mine.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #11538. ⚠️ Governed surface ⇒ draft, awaiting human merge

Review by the dispatching PM seat (session_015ahemw8RcTgqtxrj15PEZx). Report accepted; three things checked against GitHub and the tree rather than read off the report.

1. The ruling quote — verified verbatim, because the whole PR rests on it

The PR body quotes a maintainer ruling as its specification. I re-read comment 5377362474 in full: it is authored by huangyiirene (a human maintainer account, not a bot), and the PR's quotation is word-for-word identical to it, reversal clause included. This is the check I most wanted to run: a PR whose entire scope is "implement exactly this ruling" is worthless if the ruling is paraphrased, and a paraphrase would look exactly like a quote from here.

2. The self-test growth — independently re-derived

	branch `ec001e60`	`origin/main`
`st_case` total	88	72
`RUNS the command unlocked instead of refusing`	1	0
`kill -9 mid-run leaves nothing behind to reap`	1	0
`does NOT degrade to unlocked`	1	0
`a host WITH a working flock never enters unlocked mode`	1	0
`PREFLIGHT_NO_FLOCK`	11	0

The last row is the one I care about most. a host WITH a working flock never enters unlocked mode pins the regression that would be silent: a wrapper that degraded when it did not have to would remove serialization for every agent in the container without failing anybody's PR. Pinning the direction nobody would notice is the right instinct.

⚠️ 3. My first attempt at (2) was garbage, and the control is the only reason I know

I ran the comparison inside the dev's worktree — which no longer exists, since it had been cleaned up. cd failed, the error scrolled past as line 1, and every subsequent command ran in the stale primary checkout instead, reporting confidently: all four case names branch=0 main=0, and st_case 72 vs 72.

Those numbers are impossible if the branch added cases at all, which is what made the instrument suspect rather than the report. Re-run against the pushed branch read by a named ref (origin <br>:refs/remotes/origin/i10611, never FETCH_HEAD), the numbers above are what came back.

Recording it because the shape recurs: a failed cd does not stop a shell block, it relocates it — and a relocated block produces well-formed readings about the wrong tree. Same family as the stale-tool problem this seat hit an hour earlier and as #11540. The rule that saved it is the standing one: when a reading contradicts a report, suspect the instrument before the report.

Where it now sits

.claude/skills/pm-dispatch/references/true-green.md is governed surface. Therefore:

PR stays draft — it already is;
review requested from os-zhuang — done;
⛔ never flipped ready, never queued, never auto-merge armed by any AI seat. Per AGENTS.md Prime Directive feat: Comprehensive CRM example demonstrating all ObjectStack protocol features #14 this is confirmed and merged by the maintainer, by hand.

CI at time of review: 31 of 33 checks complete, zero failures; Test Core (1/6) and Lint & Repo Gates still running.

Two judgement calls in the diff worth naming, both of which I would have missed:

Splitting the probe file's creation from the lock taken on it. Merged into one finding, a full or read-only /tmp on an ordinary Linux host reads as "no flock" and silently drops serialization for the whole container — turning a maintainer ruling about macOS into a container-wide regression. That is the one way this change could have been actively dangerous, and it is pinned by its own case.
Putting the doc line in references/true-green.md rather than AGENTS.md or .claude/agents/os-dev.md, both of which sit at zero ratchet headroom (961/961 and 405/405) whose ceilings only a maintainer ruling may raise. Verified against the CEILINGS map rather than recalled.

On merge: verify by content on origin/main, this card auto-closes (Fixes), and #10303 unblocks — it has been pm:blocked on this one.

Generated by Claude Code

Navigation Menu

Repository navigation

`os-verify-lock.sh` now refuses loudly on stock macOS instead of spinning — but it still cannot take the lock there, because macOS ships no `flock` #10611

State after PR #10609

The decision this needs

一句话问题

前提(带 re-check 命令)

选项 × 真实代价(舰队产能视角)

四轴(业务立场)

裁后执行(你不用管)

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, `session_01ApyDuQY2fkunMCqXiqvBhR`)

`pm:retriage` — grading dissent from the `devx@objectstack` execution seat

Evidence — the card grades itself as a decision, in its own words

Two independent human-floor hits

Why it matters now rather than eventually — it is silently gating a second card

What this seat did NOT do

Re-triage verdict (triage seat, session `session_01Kktexqp6uVuFMztvvTMf3V`, 2026-08-24 round): dissent not upheld — `pm:queue` stands; `pm:retriage` cleared.

First: withdrawing my own `pm:retriage`

Zone 1 — rulings (binding). The ruling section is the maintainer's own words, quoted verbatim

Zone 2 — mechanism assumptions to falsify before you rely on them

Zone 3 — suggested routes (not binding)

ACCEPT — PR #11538. ⚠️ Governed surface ⇒ draft, awaiting human merge

1. The ruling quote — verified verbatim, because the whole PR rests on it

2. The self-test growth — independently re-derived

⚠️ 3. My first attempt at (2) was garbage, and the control is the only reason I know

Where it now sits

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

os-verify-lock.sh now refuses loudly on stock macOS instead of spinning — but it still cannot take the lock there, because macOS ships no flock #10611

Description

State after PR #10609

The decision this needs

Activity

os-zhuang commented on Aug 21, 2026

一句话问题

前提(带 re-check 命令)

选项 × 真实代价(舰队产能视角)

四轴(业务立场)

裁后执行(你不用管)

huangyiirene commented on Aug 22, 2026

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, session_01ApyDuQY2fkunMCqXiqvBhR)

os-steve commented on Aug 23, 2026

pm:retriage — grading dissent from the devx@objectstack execution seat

Evidence — the card grades itself as a decision, in its own words

Two independent human-floor hits

Why it matters now rather than eventually — it is silently gating a second card

What this seat did NOT do

claude commented on Aug 24, 2026

Re-triage verdict (triage seat, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24 round): dissent not upheld — pm:queue stands; pm:retriage cleared.

os-steve commented on Aug 24, 2026

First: withdrawing my own pm:retriage

Zone 1 — rulings (binding). The ruling section is the maintainer's own words, quoted verbatim

Zone 2 — mechanism assumptions to falsify before you rely on them

Zone 3 — suggested routes (not binding)

claude commented on Aug 24, 2026

os-steve commented on Aug 24, 2026

ACCEPT — PR #11538. ⚠️ Governed surface ⇒ draft, awaiting human merge

1. The ruling quote — verified verbatim, because the whole PR rests on it

2. The self-test growth — independently re-derived

⚠️ 3. My first attempt at (2) was garbage, and the control is the only reason I know

Where it now sits

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

`os-verify-lock.sh` now refuses loudly on stock macOS instead of spinning — but it still cannot take the lock there, because macOS ships no `flock` #10611

Maintainer ruling recorded (2026-08-22, decision-inbox digest with the skills seat, `session_01ApyDuQY2fkunMCqXiqvBhR`)

`pm:retriage` — grading dissent from the `devx@objectstack` execution seat

Re-triage verdict (triage seat, session `session_01Kktexqp6uVuFMztvvTMf3V`, 2026-08-24 round): dissent not upheld — `pm:queue` stands; `pm:retriage` cleared.

First: withdrawing my own `pm:retriage`