review / adversarial-review forward --model unnormalized, so the documented `spark` alias fails as "not supported when using Codex with a ChatGPT account"

[drewyd]

review and adversarial-review accept --model/-m (handleReviewCommand, valueOptions: ["base", "scope", "model", "cwd"], aliasMap: { m: "model" }), but they pass the raw string straight to executeReviewRun → runAppServerReview → thread/start. normalizeRequestedModel() — the function that resolves MODEL_ALIASES — is only ever called on the task path.

main (1.0.6), plugins/codex/scripts/codex-companion.mjs:

• L103 function normalizeRequestedModel(model) — resolves MODEL_ALIASES (spark → gpt-5.3-codex-spark)
• L773 const model = normalizeRequestedModel(options.model); — inside handleReview (task). The only call site.
• L714 valueOptions: [..., "model", ...] — handleReviewCommand accepts the flag
• L746 model: options.model, — forwarded raw, never normalized

The alias is documented for the runtime in skills/codex-cli-runtime/SKILL.md ("Map spark to --model gpt-5.3-codex-spark") and agents/codex-rescue.md, with nothing marking it as task-only, so --model spark on a review reads as supported.

Repro

node scripts/codex-companion.mjs review --wait --scope working-tree --model spark

Actual (exit 1):

[codex] Codex error: {"type":"error","status":400,"error":{"type":"invalid_request_error",
"message":"The 'spark' model is not supported when using Codex with a ChatGPT account."}}

Expected: the same run the full slug already gives (exit 0, review emitted):

node scripts/codex-companion.mjs review --wait --scope working-tree --model gpt-5.3-codex-spark

Verified 26 Aug 2026 against plugin 1.0.4 with codex-cli 0.149.1 on macOS 15 (arm64), ChatGPT sign-in. Code paths re-checked against main @ 1.0.6 — unchanged.

Why it is worth more than a one-line alias fix

The server message names the account type as the cause, which is wrong here and expensive to chase: the account is fine, the model is fine (the full slug works on the same account seconds later), and the literal string spark never was a model id. openai/codex#15648 is people landing on that same sentence. Anyone who reads the runtime skill, types the documented alias, and hits this has no path from the error text to the real cause.

Suggested fix

Call normalizeRequestedModel(options.model) in handleReviewCommand the way handleReview does, so both command families resolve aliases identically. Happy to send the PR.

Related: #654 (the flag is undocumented for these two commands), #651 / #476 (--effort has the same task-vs-review asymmetry).

[drewyd]

PR up: #688. Three lines of source — normalize in handleReviewCommand the way handleTask already does, which covers both review and adversarial-review since they share the handler.

Writing the test surfaced a second gap worth naming: the fake app-server recorded only turn/start, and review sets the model at thread/start, so no test could observe the model on the review path at all. That is why the existing task forwards model selection and reasoning effort test has no review counterpart. Added state.lastThreadStart to the fixture, then one regression test per command. Both fail on main with actual: 'spark', expected: 'gpt-5.3-codex-spark'.