[bwachman]

Note on base

This branch is stacked on the cl-10200-make-parked-worker-ask_director-questions-visible-and branch (tip 108bae6), which provides the worker-wait / parked-question machinery this feature coexists with. The diff in this PR is the CL-10189 change only.

Summary

Show the required operator input at the composer: when a primary session's ask_operator gate requires operator input, a persistent, theme-aware, width-safe INPUT REQUIRED strip renders in the composer-adjacent row with a provenance marker. The strip is keyed and deduped by OperatorGateEvent.id, only the primary ask_operator is eligible (fails closed for MCP/unmarked/permission/ask_director/OAuth), and every terminal path removes exactly the settled ID. Overlay and routing behavior are unchanged.

Verification

After the rebase onto main and the review fixes, CI passes on 8638639; the results below are from the original verification.

Gate	Result	Exit
bun test src/tui/operator-input-required.test.ts src/tui/operator-input-required-chrome.test.ts src/tui/gate-wire.test.ts src/tui/product-host.test.ts src/tui/worker-wait.test.ts src/tui/worker-wait-chrome.test.ts src/tui/geometry.test.ts src/tui/runner/session.operator-input-required.test.ts	223 pass / 0 fail, typecheck clean, no regression	0

All 7 findings in CL-10189-review-v2.md resolved (geometry invariant; gate-wire same-ID drop removed with two-distinct-objects tests; AC4 free-text/Esc/timeout route + product-host teardown + Phase-4 session integration; AC1 persistence incl. live theme switch; wrong-marker test; stale comments; plan negatives).

AC to code and test mapping

• AC1 persistent INPUT REQUIRED strip, provenance marker only in primary onOperatorGate: src/tui/runner/session.ts + src/tui/operator-input-required-chrome.test.ts (persistence incl. live theme switch)
• AC2 keyed/dedup by OperatorGateEvent.id, (+N more): src/tui/operator-input-required.ts + operator-input-required.test.ts + operator-input-required-chrome.test.ts
• AC3 only primary ask_operator eligible, fails closed: src/tui/gate-events.ts predicate + operator-input-required.test.ts + gate-wire.test.ts (MCP/unmarked/permission/ask_director/OAuth negatives)
• AC4 every terminal path removes exactly the settled ID: src/tui/gate-wire.ts settleOnce + gate-wire.test.ts (accept/text/Esc/timeout/abort/teardown) + operator-input-required.ts:77-89
• AC5 overlay/routing unchanged: src/tui/gate-wire.ts (one listener, NOOP default hooks) + product-host.test.ts
• AC6 theme-aware width-safe composer-adjacent row, collapse policy: src/tui/geometry/zones.ts + geometry.test.ts + geometry/resolve.ts + operator-input-required-chrome.test.ts

Deferred / Notes

• User-facing recovery interaction is a separate ticket; non-goals per BUILD_PLAN (no clickable strip, no persistence across restart, no changes to ask_director/send_input/permission/OAuth).

Refs CL-10189

[linear-code]

CL-10189

[TheGreatAxios]

Strip implementation is sound — fails-closed eligibility, settle-once across every terminal path, and overlay/routing invariance all verified (221/0 on the PR's own gate). Five items: one should-fix (strip not cleared on session clear), a same-id duplicate to pin, a doc overstatement, a comment omission, and a wrong AC1 test mapping in the body.

AC1 credits operator-input-required.test.ts for "persistence incl. live theme switch"; the live-switch test is in operator-input-required-chrome.test.ts:183 and the unit test is view-model-only. Fix the mapping.

[bwachman]

AC1 mapping fixed in the PR body: the live theme switch test is in operator-input-required-chrome.test.ts (operator-input-required-chrome.test.ts), and the body now credits that file for AC1. I also refreshed the stale base tip and test count in the body.