[qwen-code-review-bot]

Verified daily translation batch from scheduled run https://github.com/QwenLM/qwen-code-docs/actions/runs/39341970689.

Review and merge to publish: the main push triggers deploy-docs.yml, which rebuilds and deploys the site. Later scheduled runs pause while this PR is open, so merging it also unblocks the next one.

Files whose structure could not be verified were restored from HEAD and are not included here; the run log lists them.

[yiliang114]

Patrol-Run: qwen-docs-pr-review

Read the full diff (32 files) and checked it against the EN mirror, against upstream at the newly synced commit, and structurally — not as prose.

What I verified

• The EN mirror is byte-faithful. All 10 changed content/en/** files are identical to QwenLM/qwen-code at 5936225e (the commit last-sync.json now records), modulo the mirror's two systematic transforms: the ./ relative-link prefix and the stripped page-top language-switch line. Compared with git show 5936225e:docs/<path>, not by reading.
• No navigation risk. All 32 paths are modifications — nothing added, deleted, or renamed — so _meta.ts and the findUnlistedEntries gate in deploy-docs.yml are not in play for this batch.
• Structural integrity across all 21 changed non-EN .md files: no U+FFFD/mojibake, code fences balanced in every file, no truncated tails, no unreplaced placeholders, no dangling mid-sentence endings. 20 of the 21 match their EN counterpart exactly on heading count by level, fence count, per-fence line counts, and table row count.
• Code blocks were not mistranslated. The only fence bodies differing from EN are localized shell comments and example prompt strings (ja/users/features/sandbox.md blocks 4/7/10/11, ja/users/features/memory.md blocks 1/7/8). Commands, flags, env vars, and JSON keys are untouched.
• The last-sync.json hashes are correct, not just churned. ko for docs/users/features/channels/github.md moved to bac332f4…, which is exactly the sha256 of that file at 5936225e — and also at the previously synced 97925919, i.e. the file did not change between them, so ko was simply behind and is now caught up. ru for docs/developers/tools/shell.md moved to 74b5c9d4…, matching ja. Languages still on older hashes stay in the backlog by design and will retry.
• Semantics spot-checked on the two largest catch-ups. ja/users/features/sandbox.md: the new Linux tool-execution-sandbox section is 49 lines against EN's 49, same heading structure, and every load-bearing claim survives — /usr/bin/bwrap, Landlock ABI 3, exit 141 on closed output, config exit code 52, SystemDefaults precedence, --bare/--safe-mode, qwen sandbox --verify. zh/ja users/features/memory.md: 20 ### headings each, in the same order as EN, with the structured-recall section and /memory migrate-team present and enableStructuredRecall / QWEN_CODE_MEMORY_STRUCTURED_RECALL intact.
• ru/developers/tools/shell.md losing 19 lines is correct, not truncation. Upstream dropped the "is_background is required" contract, so the stale mandatory-parameter prose and the npm run dev & alternative example both go away, replaced by the optional-parameter wording and the "explicit & is not the managed background mechanism" note.

Non-blocking

1. ko/users/features/channels/github.md:92-93 — ko's option table carries two rows EN does not (groups.*.senders, groups.*.allowedUsers), which is why the shape check reports 11 table rows against EN's 9. The content itself is accurate: upstream documents both options in prose (docs/users/features/channels/github.md:101,103,105,109) and in the JSON example (:45,52,53), and this same diff correctly moves ko's example off the removed top-level senderPolicy onto operators plus groups["*"]. So ko is a superset of EN rather than a mistranslation. Worth converging on the EN table shape in a later batch; nothing here misinforms a reader.
2. Pre-existing and untouched by this PR: the import fences in developers/daemon-client-adapters/web-shell.md read } from ' @qwen-code/sdk/daemon'; with a stray space inside the string literal, in de/fr/ja/zh/ko/pt-BR/ru (2 occurrences each). Confirmed present on main, so it predates this batch, and en is clean. It breaks copy-paste of the import and deserves its own fix.
3. Carried over from #325 and still open: ru/users/features/channels/github.md:91-92 duplicates the reasonFilter row, and ru's recorded hash is already current, so it will not self-heal.

What I did not check

I did not read de/fr/pt-BR/ru prose for meaning — the shape and fence checks plus the ja/zh spot-checks above would not catch a semantic inversion in a language I did not open. I also did not build the site or run the translator locally; this verdict rests on the three PR checks (Translator unit tests and Website unit tests green, structural gate skipped) plus the byte-level mirror comparison.