[abruption]

Summary

Several localized operator guides treated protocol receipts as commercial receipts, confused policy permissions with native support, and mistranslated operational concepts such as replay, crash, redaction and process identity. This aligns Korean, Japanese and Simplified Chinese terminology with the English contracts and restores omitted limits without expanding supported behavior.

• Distinguish handoff confirmation/ACK information, persisted paired-request processing results, historical Control operation results and SSH host-key callback evidence. None is promoted to agent reading, completion or current device state.
• Correct concrete identity/key, fingerprint, replay/retry, crash/conflict, masking/deletion, advisory-lock and related terms. Preserve historical versions, observations and outstanding acceptance gates.
• Restore the generation-precondition subsection to its canonical position and heading depth in all three CLI guides, retaining existing anchors.
• Clarify existing English discovery/storage, SSH BatchMode and pairing-confirmation descriptions from the current implementation, then align all locales. Repair one accidental HTTP-200 list break in a historical record.
• Strengthen the documentation verifier to compare fence-free heading depth order and correctly recognize wrapped inline code without treating translated surrounding prose as a command.

Issue references

Refs #373

Type

• docs
• chore / refactor / test / perf

Coverage

The per-file audit covers all 51 existing localized docs per language (153 files): all 21 current guide/reference entries were compared in full, while 30 historical records per locale received term-context review. Changed historical records also received full-prose Flash comparison. Root READMEs and the release/security/plugin guides were compared separately. This is not a claim of a complete historical retranslation.

The PR changes 93 files: 81 localized docs (KO 25, JA 26, ZH-CN 30), three localized READMEs, two localized release guides, five narrowly clarified English sources, the canonical-source manifest and the documentation verifier. English source edits are limited to CLI/diagnostics/MCP/Control wording and the historical HTTP-200 line break.

Validation

• Python 3.13.7, macOS arm64: documentation suite 92 tests passed; git diff --check and python tools/generate_session_peer.py --check passed.
• The new heading check failed on all three misplaced CLI sections before repair. The inline-code regression uses the actual wrapped vault status example and proves surrounding translated prose is not extracted as code.
• Direct checks preserve fenced examples, link destinations and old anchors. Inline additions/selector corrections and previously omitted numeric qualifiers are explicitly checked against the English source; no historical numeric value is removed or changed.
• Frozen English handoff contract and handoff fixture remain byte-identical. Runtime implementation, version, dependency files and published artifacts do not change.
• Local core/adapter/Relay/MCP/Control runtime and distribution suites were not rerun: this changes documentation and its verifier, not those implementations. No live native/SSH/pairing/ACK or operational qualification is claimed.

User-requested independent translation review

The selected single external reviewer is Antigravity CLI with requested and init-confirmed gemini-3.8-flash-medium, medium effort, fresh non-interactive stdin calls in plan+sandbox mode, bounded to 600 seconds (650-second parent timeout). No repository tools, tests, network actions or writes were requested from the reviewer.

• Initial pass: 18 valid responses across 38 canonical documents and all three translations, 284 full-prose comparison units. Unchanged fences were checked directly and omitted; public UUIDs/source hashes/historical temp paths were redacted consistently. The three README translations retain their earlier complete-file Flash review because their bytes did not change.
• Correction pass: 7 valid responses, 60 corrected section units, 630892 input bytes. Targeted final pass: 3 valid responses, 30 units of corrected paragraphs with complete English context, 171260 input bytes. No failed call, fallback model or retry in these 28 operator-guide calls; init model matches in each. Recorded step types are only user input and agent response; no external tool-execution events were observed.
• Ordered input-hash manifest digests (SHA-256): initial fe62298e4864ed7c21a0366c8bfeff425a986489fbbca843ef238783574355b5; correction e52d1555a59896386937ad3611309f5791be57c2b987c5efc8702305b99fd7af; targeted 81444c71cefe1379f530ae6dc530d114ec29dbd8ba72e0fe2bfc4a1738487695. The inputs are frozen and retained privately with per-file source hashes and per-call metadata.
• After the targeted reviews, Codex applied and directly source/code-validated 21 small line replacements in 15 documents (including native target identity rather than OS identity, receipt/manual-confirmation input, size/time bounds and snapshot/tombstone qualifiers). These final small edits passed the same structural checks and 92 documentation tests, but were not sent for another Flash review. The final artifact is commit 92643d40a1b5f94d926d4aad433544e041fdd2c1; no exact-final-bytes model approval is claimed.
• CLI version was observed as 1.5.0 before the initial pass and 1.5.1 before finalization. Initial stream metadata exposes the model, not the CLI build, so the exact build of each initial call is not asserted. Final calls record version observations before and after each invocation. Codex did not run an install or upgrade command; the cause and timing of the observed CLI build change were not investigated.

Codex and locale agents compared suggestions with the English source and relevant implementation. Rejected proposals included removing the real trusted-channel fingerprint input, recasting SSH host-key evidence as handoff ACK, treating status lookup as blocking, and inventing stronger incarnation/storage guarantees. Valid corrections were applied; model agreement alone is not treated as contract evidence. Supplementary source/verifier review found no blocker and is distinct from Flash translation review.

Checklist

• Runtime CLI version/help: not applicable (runtime unchanged); Antigravity review options/model checked separately.
• Generated standalone checked without changing generated output.
• Commits follow Conventional Commits.
• Relevant English and translated documentation aligned; canonical manifest refreshed.
• No secrets, tokens or session keys added.

Current-head remote checks

Final head: 92643d40a1b5f94d926d4aad433544e041fdd2c1.

• 43/43 checks successful, including CI run 41993415553, the release gate, CodeQL and GitGuardian.
• Codex code re-review completed for this exact head; has_any_findings=false and has_unresolved_findings=false, with no inline review comments. The earlier README-only review is not substituted.
• Security-review metadata still references the earlier ff70c2c revision with null finding flags. It is not described as a fresh security review of this head or as a security certification.

The PR remains open for owner review. No merge, release, installation or operational change was performed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-05T08:21:51.512468Z	92643d4	Manual request
🔒 Security Review	✅ Completed 2026-11-05T06:38:51.157678Z	ff70c2c	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[abruption]

@codex review

Please review the expanded current head 40864f0d947262def4f261caa15e3e677d184d28, now covering Korean, Japanese and Simplified Chinese README terminology. The initial review covers only the earlier Korean-only head ff70c2c; English/runtime remain unchanged.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: 40864f0d94

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[abruption]

@codex review

Please review current head 92643d40a1b5f94d926d4aad433544e041fdd2c1. This revision expands the README-only review to operator terminology in Korean, Japanese and Simplified Chinese, narrow canonical English clarifications, and the documentation verifier. The earlier 40864f0 review does not cover this expansion. Runtime implementation and the frozen English handoff contract/fixture are unchanged. The PR body records source evidence, translation-review coverage and final direct-validation limits.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: 92643d40a1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".