fix(review): accept positional focus text for parity with adversarial-review

[axisrow]

What

Removes the unconditional focus-text rejection in /codex:review so it no longer aborts where /codex:adversarial-review succeeds.

Closes #522.

Why

/codex:review rejected any positional argument as "custom focus text" (validateNativeReviewRequest), while /codex:adversarial-review accepts it. Two sibling review commands behaved differently in a way the user only discovered by hitting the error, and the suggested escape hatch (adversarial-review) changes the review semantics — not a substitute for "let me name the model in my own words."

This also blocks non-English model/effort entry: even when a host (e.g. Claude Code reading the command's .md) rewrites a natural-language phrase into --model/--effort, any residual positional word still aborted the native review (see the Japanese example in #522).

Change

Remove the focus-text branch from validateNativeReviewRequest. The native reviewer (review/start) does not consume focus text, so leftover positional words are now silently ignored rather than rejecting the invocation. Review semantics are unchanged (still native structured-review via review/start); only the spurious abort is gone.

This PR does not add language/alias recognition in the companion — that stays the host's job. The fix is purely about interface parity: review no longer fails where adversarial-review succeeds.

Scope

This is independent of #471 — the two touch different concerns (#471 is model/effort negotiation + broker refresh; this is a review-command parity fix). Based on main so it can land on its own.

Testing

• New runtime test review accepts (ignores) positional focus text for parity with adversarial-review (red before, green after).
• Updated the stale review rejects focus text test and the commands.test.mjs assertion that pinned the old wording.
• 0 regressions vs main (the pre-existing suite failures are unchanged).

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a9b9481e97

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".

[chatgpt-codex-connector]

Reject unknown switches before ignoring focus text

Because parseArgs treats any unrecognized --... token as a positional, ignoring all focusText here also silently accepts typos or unsupported flags such as /codex:review --bas main or /codex:review --scoope branch; in those cases the command now runs a review against the auto-selected target instead of failing, which can give users a successful review of the wrong changes. Please distinguish real positional focus text from option-looking positionals before dropping it.

Useful? React with 👍 / 👎.

[axisrow]

🔍 Local review (cycle 1)

Reviewed locally (Claude subagent + Codex companion), no bots pinged. Critical-only bar.

Verdict	Reviewer	Finding	Location
SKIP	claude	Dead focusText local + vestigial _focusText param in the native-review branch (computed L366, passed to validator L369, used nowhere now)	codex-companion.mjs:366,369

Codex companion: approve, 0 findings — "no critical security, data-loss, correctness, or performance defect is supported by the PR diff."

Claude subagent: 0 critical findings after verifying 6 claims against the code (focusText cannot leak into review/start RPC — payload is exactly {threadId, delivery, target}; cannot reach job metadata or rendered output; staged-only rejection still enforced; test is a genuine inversion, not a weakened assertion; parity-only scoping is sound).

No FIX verdicts → final cycle. The one SKIP (dead param) will be addressed in the cleanup pass or left as-is since it's harmless.

[axisrow]

🔍 Local review (cycle 1, round 2 after cleanup commit)

Reviewed locally (Claude subagent + Codex companion), no bots pinged. Critical-only bar.

Verdict	Reviewer	Finding	Location
SKIP	claude	validateRequest?.(target, focusText) still passes a now-dead 2nd arg (validator takes 1). Harmless — JS drops it. One-line cleanup.	codex-companion.mjs:729

Codex companion: approve, 0 findings — "no defensible critical security, data-loss, correctness, or performance risk."

Claude subagent: 0 critical after verifying 6 claims against the code (native path never reads focusText; cleanup commit did NOT break the adversarial path — focusText still reaches buildAdversarialReviewPrompt; no leak into prompt/job metadata/logs; review --scope working-tree <leftover> runs native exit 0 without passing leftovers as instructions; tests are genuine inversions not weakened; pre-existing failures confirmed identical on main).

No FIX verdicts → final cycle. The one SKIP (dead 2nd arg) applied in the cleanup pass.