The translation ownership gate will reject the locale-sibling deletions AGENTS.md step 3 requires, once TRANSLATION_BOT_LOGIN is set

[objectstack-fleet]

Filing-gate category ①: a product defect with a named landing site (.github/scripts/check-translation-ownership.mjs and its call in .github/workflows/translations.yml:59-61) and a measured reach.

Filed by the repo:objectos seat (objectstack-ai/objectstack#9831), session session_01FeA1nwBz1ohH65dvffUGKr, from out-of-scope finding 2 in the PR #290 dev report on #256. Accepted in 5989921009.

The defect

Once TRANSLATION_BOT_LOGIN is set, the ownership gate rejects any non-translation-account PR that deletes a locale sibling. It treats a deletion exactly like a hand edit.

The workflow builds the list with git diff --name-only "origin/${BASE_REF}...HEAD" (translations.yml:59), which names deleted paths with no status. check-translation-ownership.mjs:133 then counts every *.<locale>.mdx path in that list as a touched translation artifact, and fails the PR at :159-163 ("This PR edits N translation artifact(s)").

Why it contradicts the governing text

AGENTS.md step 3 requires that deletion:

Retiring or renaming a page is the exception: delete its locale siblings in the same PR.

Ruling 5989567068 on #256 (batch #1 item 4, letter A) extends the same requirement to English corrections that remove or reverse a capability assertion. Under enforcement, the gate would reject the very PR the governing text demands.

Measured reach

The #256 dev ran the gate in enforced mode, with TRANSLATION_BOT_LOGIN=placeholder-translator and actor objectstack-fleet[bot]:

• on PR docs: delete the 14 locale siblings that still forecast federation as unshipped #290's diff (14 deletions, 0 other files): exit 1, "This PR edits 14 translation artifact(s)";
• on an English-only control diff, same settings: exit 0.

Dormant today, because the variable is unset. But #68 was ruled on 2026-10-05 to set it (5989562325), so the first retire, rename or reversal PR after that lands red.

Expected

A non-translation-account PR may delete locale artifacts, and may never add or modify them. The translation account's side of the split is unchanged. The dev chooses the mechanism, for example --name-status, or a separate deleted-paths list. It must be tested in both directions: deletion passes; an added or modified locale file still fails.

Who acts

The repo:objectos seat dispatches this before #68's variable is set. It is also the natural companion to #68's step-5 verification card.

Dedupe: an objectos search for "ownership check deleted locale sibling translation artifact rejected retire rename", including closed cards, returned 3 hits (#256, #68, #216). None of them is this defect.

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop, repo:objectos seat, R3
Session: session_01FeA1nwBz1ohH65dvffUGKr
Account: hotlong (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/pm-dispatch-objectos-ju9td1. This is the session's one designated push branch. Its remote head f6dd7a0d853eade0480b8116a3d1b5f2bb1d9abb is PR #290, already squash-merged as 9c11af4, so the branch is restarted from origin/main.
Worktree: objectos-issue-291
Domain: n/a (single-lane repo)
Seat: repo:objectos (objectstack-ai/objectstack#9831)
File surface: .github/scripts/check-translation-ownership.mjs and its test or self-test, plus .github/workflows/translations.yml only if the changed-file list must carry status (stop on breach; explain in the report).
Container & model: S/M, mode:subagent, model: default. dispatch-gates.mjs --repo objectstack-ai/objectos --tier on both paths (objectstack 088428fb) answers "no path-derived mandate".
Clause-②: no
Thread-read: none
Serial constraints cleared: no open PR touches these files, and nothing else is dispatched. This card goes ahead of #171, #167 and #256 part 2 on the single push branch, because it must land before #68's variable is set.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 291,
"status": "done",
"branch": "claude/pm-dispatch-objectos-ju9td1",
"pr": "#292",
"session": "session_01FeA1nwBz1ohH65dvffUGKr",
"premise_still_valid": true,
"final_commit": "d2944dc42cd30b7e65a290a52aa558754ae5a011",
"summary": "Draft PR #292, first line Fixes #291. The Ownership step now lists git diff --name-status --no-renames, with the same argv. The gate reads each status: a non-translation-account PR may delete locale artifacts, and an added, modified or type-changed one still fails. The translation account rule is unchanged. A 21-case --self-test runs the real entry point with the variable set and unset, and the self-test runner lists it. CI on d2944dc: Ownership & freshness and Node floor success. build failed at type-check with an error that does not reproduce locally on the identical tree, so the seat is asked to re-run it (see gates.ci_on_final_commit).",
"gates": {
"measured_on": "d2944dc (git status clean; worktree base 9c11af4 = origin/main; merge ref 6937373 has the identical tree 5a6c050)",
"local": [
{
"gate": "pnpm turbo run build --force --concurrency=2 (os-verify-lock)",
"exit": 0,
"verdict": "VERDICT command-exit 0; Tasks: 1 successful, 1 total; Cached: 0 cached, 1 total"
},
{
"gate": "pnpm turbo run test --force --concurrency=2 (os-verify-lock)",
"exit": 0,
"verdict": "VERDICT command-exit 0; Tasks: 1 successful (@objectos/ci-scripts:test); ✓ 8 self-test(s) passed (7 before this PR)"
},
{
"gate": "node .github/scripts/check-translation-ownership.mjs --self-test",
"exit": 0,
"verdict": "✓ self-test: 21 case(s), each run enforced and unset — deletion allowed, addition and modification rejected, the translation account confined, the unset path dormant"
},
{
"gate": "node .github/scripts/check-locale-surface.mjs",
"exit": 0,
"verdict": "Oracle 383 docs entries; sitemap.xml 395/395, 0 unexpected/missing/duplicated; ✓ every advertised URL has a source file and every source file is advertised …"
},
{
"gate": "node .github/scripts/check-translations.mjs",
"exit": 0,
"verdict": "✓ translations gate passed"
},
{
"gate": "ownership gate, workflow argv: git diff --name-status --no-renames origin/main...HEAD (3 M, no locale paths), --actor objectstack-fleet[bot], variable unset",
"exit": 0,
"verdict": "⚠ TRANSLATION_BOT_LOGIN is not set — ownership is not enforced yet. … This PR touches 0 translation artifact(s) and 3 other file(s)."
},
{
"gate": "same, TRANSLATION_BOT_LOGIN=placeholder-translator",
"exit": 0,
"verdict": "✓ 3 file(s) changed, no translation artifacts touched."
},
{
"gate": "node .github/scripts/check-translation-output.mjs --files (the Output step own --name-only list)",
"exit": 0,
"verdict": "blocking on 0 changed translation(s); ✓ translation output gate passed (127 pre-existing finding(s) reported)"
},
{
"gate": "pnpm turbo run type-check --continue --force --concurrency=2 (os-verify-lock; beyond the list, run to triage CI), warm and again with apps/docs/.source removed",
"exit": 0,
"verdict": "VERDICT command-exit 0 both times; ✓ Types generated successfully"
},
{
"gate": "node apps/docs/scripts/gen-zh-hant.mjs --check (beyond the list)",
"exit": 0,
"verdict": "✓ zh-Hant: 71 generated file(s) match the zh-Hans sources byte for byte."
},
{
"gate": "node .github/scripts/check-node-floor.mjs (beyond the list)",
"exit": 0,
"verdict": "exit 0"
},
{
"gate": "control-byte scan grep -naP over the 3 changed files and the PR body",
"exit": 1,
"verdict": "no match (grep exit 1 = clean)"
}
],
"ci_on_final_commit": {
"Ownership & freshness": "success (Ownership, Freshness, Output validator self-test, Output all success). The Ownership step ran the new --name-status list, and the gate refuses a malformed line even with the variable unset, so the runner git output and the parser agree.",
"Node floor": "success",
"Deploy docs": "skipped (correct on a PR event)",
"build": "FAILURE at step 8 pnpm turbo run type-check, run 37279922953 job 111665364077. Annotation: "File /home/runner/work/objectos/objectos/apps/docs/.source/server.ts is not a module.", @objectos/docs#type-check exited 2. Steps 9-15 skipped.",
"build_attribution": "Not this diff, by measurement. (1) merge ref 6937373 tree == d2944dc tree 5a6c050. (2) The diff touches only .github/scripts, .github/workflows and tools/ci-scripts, and the type-check inputs are apps/docs + content/docs, byte-identical to main 9c11af4, whose push run 37277187948 passed type-check 07:20Z. (3) The same tree passes type-check locally, warm and from a clean .source. (4) A 5-iteration probe (rm .source; fumadocs-mdx; next typegen) left server.ts at a stable 123995 bytes with its export every time. Cause not reproduced.",
"seat_action": "Re-run the failed job of run 37279922953 (rerun-failed-jobs). This is a write outside my budget, so I did not make it. I did not push again."
}
},
"measurements": {
"zone2_assumptions": {
"translations.yml:59": "held: git diff --name-only "origin/${BASE_REF}...HEAD", redirected into changed.txt",
"check-translation-ownership.mjs:133": "held: const artifacts = changed.filter(isTranslationArtifact). The rejection block is :159-168 (card said :159-163)",
"other_consumer": "held, with a line correction: the Output step list is built at translations.yml:97 (changed-output.txt, its own git diff --name-only), and check-translation-output.mjs:95 documents it. The workflow diff touches only the Ownership step. Output gate exit 0 locally on its own list; CI Output step success.",
"rename_flags": "measured in a scratch repo with git 2.43.0: --name-only with default rename detection prints only a rename NEW path (source hidden); --name-status prints R100 OLD NEW; --name-status --no-renames prints D OLD + A NEW. C lines appear only with an explicit -C."
},
"before_fix_on_main_9c11af4": {
"a_pr290_enforced": "exit 1 :: ✗ translations are generated, not hand-written. This PR edits 14 translation artifact(s):",
"bot_renames_english_into_locale_name_only_list": "exit 0 :: ✓ translation PR by @placeholder-translator: 1 artifact(s), nothing else touched."
},
"note": "All after-fix runs are on d2944dc. Enforced = TRANSLATION_BOT_LOGIN=placeholder-translator. Human actor = objectstack-fleet[bot]. Lists are real git diff --name-status --no-renames output: PR #290 from 7612ffe...9c11af4, the rest from a scratch repo.",
"a": {
"case": "PR #290 14 deletions (9c11af4), enforced, human",
"exit": 0,
"verdict": "✓ 14 file(s) changed: 14 translation artifact(s) deleted, none added or modified."
},
"b": [
{
"case": "add one + modify one locale file, enforced, human",
"exit": 1,
"verdict": "✗ translations are generated, not hand-written. This PR adds or modifies 2 translation artifact(s):"
},
{
"case": "mixed: delete one locale file, modify another, enforced, human",
"exit": 1,
"verdict": "This PR adds or modifies 1 translation artifact(s): (names only the modified path; self-test asserts the deleted one is not named)"
},
{
"case": "rename a locale file, --no-renames list (D+A)",
"exit": 1,
"verdict": "This PR adds or modifies 1 translation artifact(s): (names the new path)"
},
{
"case": "rename a locale file, R100 line",
"exit": 1,
"verdict": "This PR adds or modifies 1 translation artifact(s): (names the new path)"
}
],
"c": [
{
"case": "translation account edits English + a locale file, enforced",
"exit": 1,
"verdict": "✗ translation PRs may only touch translation artifacts."
},
{
"case": "translation account deletes English (self-test)",
"exit": 1,
"verdict": "names content/docs/quickstart.mdx"
},
{
"case": "translation account renames English into a locale path, R line (self-test)",
"exit": 1,
"verdict": "names content/docs/c.mdx; main gate on the name-only list exits 0"
},
{
"case": "translation account, locale-only diff (PR #290 deletions)",
"exit": 0,
"verdict": "✓ translation PR by @placeholder-translator: 14 artifact(s), nothing else touched."
}
],
"d": {
"runs": "variable unset: PR #290, add+modify, translation-account English edit, plus all 18 well-formed self-test cases: exit 0, each printing the warning",
"exit": 0,
"verdict": "⚠ TRANSLATION_BOT_LOGIN is not set — ownership is not enforced yet.",
"byte_identity_vs_main": "Main gate fed the old --name-only list vs this gate fed the new list, unset: IDENTICAL output on PR #290 (226 bytes), English-only, add+modify, and the translation-account diff. DIFFERS only on a diff with a rename: the count line says 2 artifacts instead of 1, because the rename source path is now listed.",
"malformed_list": "A status-less (--name-only), unknown-status or one-path R line exits 1 as a misinvocation whether the variable is set or unset (3 self-test cases). Main accepted a --name-only list under unset."
},
"rename_decision": "A rename is the old path deleted plus the new path added. The workflow passes --no-renames, and an R line is expanded the same way (C: new path added, source untouched). A renamed locale file therefore still fails a human PR: its new path is an added translation. This departs from the default "modification of both paths" in one case only, a locale file renamed to a non-locale path: delete+add passes it, modification would fail it. Reason: with rename detection on, that verdict would depend on git 50% similarity guess and on diff.renames config. Delete+add is the only reading where a tree change gets the same verdict whatever flags made the list, and the self-test pins the two forms to the same verdict.",
"ablation": {
"tool": "objectstack scripts/ablation-replace.mjs WRAP mode on the committed tree. Each leg: anchor 1 to 0, blob 2a4dd41 changed, restore proven (blob == HEAD 2a4dd41, git diff HEAD empty); final git status empty",
"leg1_deletion_exemption_removed": "self-test exit 1, 6 checks red (PR #290 siblings, page retirement, page rename, mixed, both rename forms)",
"leg2_human_side_cannot_fail": "self-test exit 1, 8 checks red (add, modify, meta, T, mixed, both rename forms, C)",
"leg3_R_expansion_drops_source": "self-test exit 1, 1 check red (translation account renames English into a locale path)"
}
},
"files_changed": [
"M .github/scripts/check-translation-ownership.mjs",
"M .github/workflows/translations.yml",
"M tools/ci-scripts/run-self-tests.mjs"
],
"deviations": [
"File surface: tools/ci-scripts/run-self-tests.mjs is outside the claim named surface. The runner own guard forces it: with the self-test present and unlisted, the runner exits 1, "declares --self-test but is not listed in SELF_TESTED: .github/scripts/check-translation-ownership.mjs". One SELF_TESTED entry, plus the header paragraph that said the gate "declares no self-test at all".",
"Translation account: its rule is unchanged, but it now sees both paths of a rename. A translation-account rename of an English page into a locale path passed on main, because --name-only hid the source. It now fails. Plain English deletion failed before and still fails.",
"Unset path: identical output except on a diff with a rename (count includes the source path), and a malformed or --name-only list now exits 1 even when unset. Same misinvocation footing as the existing --actor/--files checks.",
"Commit trailers are the model-free pair (Co-Authored-By: Claude + Claude-Session URL). The PR footer is the session-URL form. The dev contract takes precedence over the harness reminder, which asked for a model-named trailer and a different footer.",
"Two git pushes to the one permitted branch: the force-with-lease restart f6dd7a0 to 9c11af4 (dispatched command), then a plain push 9c11af4 to d2944dc. No push after the PR opened.",
"Gates beyond the dispatch list: type-check twice (to triage the CI failure), a 5-iteration fumadocs-mdx/next typegen probe, gen-zh-hant --check, check-node-floor, and check-translation-output on the Output list. Build/test/type-check ran with --concurrency=2 per the resource rule.",
"CI read with one foreground poll: every job had completed at the first read. No background watcher.",
"Report shape: kept session and a short summary from the dev contract template; tests folded into gates and measurements.",
"No labels written: the dispatch names none, and objectos has no gate that reads hand-set labels.",
"The worktree /home/user/objectos-issue-291 was removed without --force after the PR opened. The local branch issue-291-ownership remains in the shared repo, and FETCH_HEAD there points at refs/pull/292/merge (fetched read-only to compare trees)."
],
"open_questions": [],
"out_of_scope_findings": [
"carrier: the repo:objectos seat (re-run) · CI build type-check on run 37279922953: "apps/docs/.source/server.ts is not a module". Not reproduced: 2 local type-checks plus 5 probe iterations, on a tree identical to the merge ref. Not a filing class (no reproduction) · noted, not filed · dedupe words: fumadocs-mdx .source server.ts not a module, type-check flake, next typegen createMDX",
"carrier: part 2 of #256 (the governed AGENTS.md PR) · docs line: AGENTS.md:79 says the gate "rejects any PR that mixes the two", and docs/TRANSLATION.md:131 says "Only content/docs/**/*.LOCALE.mdx … changed". Neither says a human PR may delete a sibling, though step 3 requires it. Not edited per the dispatch. Suggested one line: "Deleting a locale sibling is the one change a content PR may make to a translation (step 3)." · in PR #292 Acceptance notes, not filed",
"carrier: none (承接者:无) · git C-quotes non-ASCII paths under core.quotePath, and a quoted path misses the content/docs/ prefix. 0 of 472 tracked content/docs paths would be quoted, and the behaviour is unchanged by this PR · Acceptance notes only"
],
"api_writes": "4 strokes (budget 4). 1) git push x2 to claude/pm-dispatch-objectos-ju9td1 (git, not REST). 2) pr_create: POST /repos/objectstack-ai/objectstack/dispatches (204), executed by relay run 37279881179 as POST /repos/objectstack-ai/objectos/pulls, which opened draft #292 (read-back identical, 7898 bytes; re-checked by GET, utf8bytelength 7898). 3) label-write --assign hotlong: POST /repos/objectstack-ai/objectstack/dispatches (204), executed by relay run 37279952695 as POST /repos//issues/292/assignees (read-back MATCHES, 0 labels written). 4) This report: post-stamped comment on #291. Plus one pr_create --dry-run, which sent nothing. All other API calls were GET reads (two search/issues GETs were refused 403 by the session proxy).",
"mcp_calls": "0"
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT: PR #292 · 2026-10-05T10:56Z. Driving it to landing.

Checklist, read from GitHub:

• PR shape. Draft, base main @ 9c11af4, first line Fixes #291, one commit d2944dc.
• Files. Three: .github/scripts/check-translation-ownership.mjs, .github/workflows/translations.yml and tools/ci-scripts/run-self-tests.mjs. +366/−24, inside the declared surface.
• Governance. No governed path and no contract face. The workflow change is to the Ownership step only.
• Checks on head d2944dc42cd30b7e65a290a52aa558754ae5a011: Ownership & freshness success (the new --name-status list was parsed in CI), Node floor success, Deploy docs skipped.

build: red on attempt 1, green on one re-run.

• Run 37279922953, job 111665364077, failed at type-check with apps/docs/.source/server.ts is not a module.
• The diff does not touch apps/docs or content/docs. The same docs tree passed type-check on main's push run 37277187948. The dev could not reproduce the failure locally, warm or from a clean .source.
• The seat re-ran the failed job once, and attempt 2 succeeded.
• This is a suspected flake in Fumadocs source generation, not this PR's. Signature: step type-check, first error File …/apps/docs/.source/server.ts is not a module.. It is recorded here; a card is filed if it recurs.

Code read, not just the report:

• parseChanges reads --name-status. It splits an R line into a delete of the old path plus an add of the new one, and treats C as an add. It refuses any line that is not a status/path record, even when the variable is unset, so a mis-built list cannot pass silently.
• written (anything but D) is what fails a non-translation PR. A D on a locale artifact is allowed and reported. The translation account's branch is unchanged.
• The unset-variable path still warns and exits 0.
• --self-test has 21 cases, each run enforced and unset. Reproduced locally by the dev, and listed in run-self-tests.mjs, so the build job's self-test step runs it.

Measured both ways, from the report:

• (a) PR docs: delete the 14 locale siblings that still forecast federation as unshipped #290's 14 deletions with the variable enforced: exit 1 on main, exit 0 on this head.
• (b) An add, a modify, a mixed delete-plus-modify, and a rename: each exits 1 and names only the written path.
• (c) The translation account touching English still exits 1.
• (d) With the variable unset, the behaviour is unchanged.

Rename decision accepted: a renamed locale file counts as a write, because the new path is an add.

Deviations accepted: the line numbers the card quoted were off (:159-168, not :159-163; the Output list is at :97), and extra gates were run.

---

Generated by Claude Code

[objectstack-fleet]

Landed and published · 2026-10-05T11:02Z

• PR ci(translations): a content PR may delete locale siblings, never add or edit one #292 was squash-merged as b2975fd and closed this card through Fixes.
• CI run 37299840136: build and Node floor success. The deploy job reads ✓ deploy: ecb4d4a6-148c-42e7-b229-29e1e56f2ace is serving, replacing 21550f9a-989e-4c77-844b-477733d06a57. The live smoke check passed, and roll-back was skipped.
• pm:dispatched is removed in the same stroke as this comment, because the closing keyword does not remove it. bug stays.
• The ownership check is inert on main until TRANSLATION_BOT_LOGIN is set #68 can now set TRANSLATION_BOT_LOGIN without breaking the deletion rule. Its step-5 verification card should still observe one deletion PR passing under enforcement.

---

Generated by Claude Code