Add deference hook phrases for steer handoff wording

Context

A live operator side quest on 2026-06-23 identified two additional deference-register phrases that should be caught by the autonomous-turn Stop-hook phrase matcher:

• Your steer on
• if you'd rather

The current matcher surface is ai/scripts/lifecycle/deferencePhraseMatch.mjs. Its DEFERENCE_PHRASES list is intentionally tight and fixture-backed; it already excludes broad near-misses like should I, shall I, happy to, no rush, and whenever you want. These two additions are narrower than those broad forms and map to the same handoff-of-maintainer-judgment pattern the hook is meant to mirror.

Release classification: boardless / post-release tooling hardening. This is not a release blocker.

Live latest-open sweep: checked the latest 20 open issues at 2026-06-23T00:46:32Z; no equivalent open ticket found. A2A in-flight sweep: checked latest 30 messages at the same boundary; no competing lane-claim or lane-intent for this phrase addition. KB ticket sweep for deferencePhraseMatch DEFERENCE_PHRASES Your steer on if you'd rather found no equivalent ticket. Exact local sweep found the historical deference-hook origin surfaces and prior real prose containing if you'd rather, but no open duplicate for adding these fixtures.

The Problem

The deference hook currently catches phrases such as would you like me to, unless you want me, want me to, do you want me, your call, and your move. Recent live coordination produced adjacent wording that has the same behavior at an autonomous turn terminal:

• Your steer on asks the operator to decide a lane that the agent should classify or route.
• if you'd rather makes the agent's next action conditional on operator preference when peer/lane evidence should decide.

Without adding these fixtures, the hook remains blind to two observed forms of the same Helpful Assistant regression register.

The Architectural Reality

ai/scripts/lifecycle/deferencePhraseMatch.mjs owns the shared phrase list, phrase matching, and reminder text consumed by lifecycle Stop hooks. The helper uses case-insensitive boundary matching, so the new phrases should be added as literal lower/upper-case strings in DEFERENCE_PHRASES and covered by the existing focused unit-test surface.

The existing JSDoc states that broadening the list requires a falsifier-backed follow-up so the hook stays a mirror rather than a noisy leash. This ticket is that follow-up: two concrete phrases, not a generic widening of all conditional or polite language.

The Fix

Add the following strings to DEFERENCE_PHRASES in ai/scripts/lifecycle/deferencePhraseMatch.mjs:

• Your steer on
• if you'd rather

Update the focused hook tests under test/playwright/unit/hooks/ so both phrases are covered by the matcher/autonomous-turn behavior without changing the operator-dialogue carve.

Contract Ledger

Decision Record impact

None. This aligns with the existing deference-register hook substrate from #13674 and does not amend an ADR.

Acceptance Criteria

• DEFERENCE_PHRASES contains Your steer on.
• DEFERENCE_PHRASES contains if you'd rather.
• Focused unit coverage proves both phrases match through the shared matcher.
• Focused Stop-hook coverage proves the phrases follow the existing autonomous-turn block path, or the PR body explains why shared matcher coverage is sufficient.
• Existing broad near-miss exclusions remain unchanged unless the PR gives new falsifier evidence.

Out of Scope

• No generic expansion to all conditional phrases.
• No changes to the operator-dialogue carve.
• No changes to lane-state parsing or no-hold state semantics.
• No new hook architecture.

Related

Related: #13674

Retrieval Hint: deferencePhraseMatch DEFERENCE_PHRASES Your steer on if you'd rather helpful assistant deference hook