docs(cli-reference): list the manifest status values in the status field

[ydflow]

Summary

Follow-up to #850. lizhengfeng101 closed that PR on direction grounds but noted the accuracy fixes found there were worth carrying forward as a small focused PR. This is the one of the four that is still wrong on main.

The status field table in the CLI reference listed only success, completed_with_warnings, completed_with_errors, and skipped. Those four values are the pre-manifest path (outputJSON / outputJSONNoFiles). Every review run that produces a run manifest takes a different branch instead — cmd/opencodereview/output.go:386 sets payload.Status = string(manifest.TerminalState) — and internal/session/manifest.go:166 defines those states as complete, partial, failed, and skipped. So the documented enum was missing exactly the values a real ocr review --format json run emits, and a consumer validating against the documented set would reject a successful run's "complete".

The change documents both paths, keeps skipped listed once with a note that it covers both the no-items-selected manifest state and the no-supported-files case (outputJSONNoFiles), and syncs the row across all five locales (en, ja, ko, ru, zh) — the same pattern used by #1555.

The other three corrections from #850

Checked against main; all three are already correct in the docs, so there is nothing to carry over:

• retry_codes only accepts 4xx — pages/src/content/docs/en/configuration.md:277 already says "Only 4xx HTTP status codes are accepted" and that 5xx cannot be added. Matches sanitizeRetryCodes in internal/llm/resolver.go:970, which errors on anything outside 400–499 and warns on 408/409/429.
• ocr session comments uses --json — cli-reference.md:442 already documents ocr session comments --json <session-id>. Matches the flag registration in cmd/opencodereview/session_cmd.go:207.
• Exit-code semantics — the exit-codes section already states non-zero only on fatal error, and faq.md:234 spells out that partial failures exit 0. Matches reviewResultError in cmd/opencodereview/review_cmd.go:326.

Verification

• make english-check — exit 0, "No unapproved non-English text in 635 scanned source files."
• Confirmed LF line endings (no CRLF) and UTF-8 on all five files, per the AGENTS.md line-ending rule.
• Confirmed the replaced rows keep the table structure: each is a 3-cell row, identical pipe count across all five locales.
• pages/ vitest and typecheck were not run: pages/node_modules is absent in this checkout and I did not install dependencies. The change is Markdown table prose only — no .ts/.tsx and no anchors or cross-references touched — so it is outside those suites' coverage. Flagging it as unverified rather than claiming a pass.

Disclosure: I used an AI assistant to draft this PR description and to locate the source lines; every claim above was verified by reading the referenced code and running english-check myself.

[github-actions]

✅ OpenCodeReview: Review skipped: no items were selected.

[lizhengfeng101]

Reviewed against the code and the fix itself looks accurate: cmd/opencodereview/output.go:385-401 sets payload.Status = string(manifest.TerminalState) whenever a manifest is present, completed_with_warnings/completed_with_errors are only reachable in the manifest == nil branch, the terminal states in internal/session/manifest.go:166-169 are exactly complete/partial/failed/skipped, and outputJSONNoFiles emits skipped for the no-supported-files case. So the old enum really would make a validating consumer reject a successful run's "complete".

One wording concern, with a suggestion on each of the five locale files: "pre-manifest path" is internal jargon that is not defined anywhere in the docs, so a reader still cannot tell which set of values their run emits. The payload embeds the manifest field itself (json:"manifest,omitempty"), and ocr review nearly always produces one — the nil-manifest branch is only reachable on manifest construction failure or the scan path (see the RunManifest comment in internal/agent/agent.go). Keying the row on the observable presence of the manifest field makes it self-contained and jq-verifiable.

Non-blocking follow-up: the canonical JSON example above this table (all five locales) still shows "status": "success" with no manifest field, which is the rare path for a real review run. Worth considering in a separate PR.

Disclosure: I used an AI assistant to help draft this review; the code references above were verified against the repository.

[lizhengfeng101]

LGTM