[codex] Fix creative transformer follow-ups

[bokelley]

Fixes the creative transformer follow-ups from PR #5219 by aligning migration/release docs with applies_to_output_format_ids, making list_transformers no-match brief expansion return no options, and validating plural target_format_ids against transformer outputs.

Adds focused regression coverage plus an empty changeset for the docs/training-agent-only patch.

Validation: focused transformer tests, typecheck, precommit, pre-push storyboard matrices, and docs broken-link check.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
adcp	🟢 Ready	View Preview	Jun 6, 2026, 12:48 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[aao-release-bot]

Clean follow-up cleanup from #5219 — docs now match the schema instead of contradicting it, and the plural-target path is no longer a validation gap. The right shape: the migration guide was the last doc still claiming per-output pricing "does not survive the move," and the schema (applies_to_output_format_ids + UNPRICEABLE_OUTPUT) already disagrees.

Things I checked

• Docs match the wire, verbatim. New prose at creative-transformers.mdx:58 ("unscoped option is the default for any output," reject with UNPRICEABLE_OUTPUT rather than guess) is a near-verbatim restatement of the authoritative descriptions in static/schemas/source/core/vendor-pricing-option.json:17 and static/schemas/source/enums/error-code.json:118. ad-tech-protocol-expert: sound. No spec drift — both fields already shipped, this PR only aligns prose.
• No dangling "split into per-price-point transformers" guidance. Both rewritten passages (L58, L70) are the only occurrences; bullet #2's "See the pricing note above" cross-ref still resolves. docs-expert: lands-well.
• Plural target validation is correct. targetIds (task-handlers.ts:6755) is the canonical merge of target_format_ids plural / target_format_id singular, capped at 50. The new findIndex maps invalidTargetIndex to the right target_format_ids[i] slot, and the req.target_format_ids?.length guard picks the right field name. Behavior change is correct: a target outside the transformer's output_format_ids now fails closed regardless of which field carried it.
• Empty-options change is fail-closed, not a regression. task-handlers.ts:3407 dropping the filtered.length > 0 guard means a zero-match brief returns options: [] instead of silently widening to the full catalog. Paging handles the empty set cleanly (moreOptions = false, no cursor). Tested. The field outlives its data was the old failure mode; this kills it.
• Empty changeset is correct, not missing. Per .agents/playbook.md:289-293, empty frontmatter is the sanctioned mechanism for changes that don't touch the published protocol surface (static/schemas/source/**). This PR touches none. code-reviewer flagged it Blocker; overruling on the strength of the playbook policy plus zero schema-source files in the diff.

Follow-ups (non-blocking — file as issues)

• task-handlers.ts:6794 — targetIds[0] ?? req.target_format_id ?? ...: the ?? req.target_format_id is dead. targetIds[0] already derives from req.target_format_id in every path, so the fallback can never change the result. Drop it.
• creative-transformers.mdx:24 — the "What changed" table still describes pricing as "per transformer" without noting per-output scoping survives via applies_to_output_format_ids. Untouched by this PR and not contradictory, just coarse. Optional one-line append.

Minor nits (non-blocking)

1. Error-message wording shift. task-handlers.ts:6785 moved from target_format_id "X" is not... to Target format "X" is not.... The new test matches on code/field only, so nothing breaks — but any downstream string-match on the old message would. Heads-up, not a fix.

Approving on the strength of the protocol-conformance verdict plus the fail-closed validation. Follow-ups noted.