[raiseCatError]

An echo-visible ownership lookup can replay an immediate answer and Enter into command approval before a macOS line reader is recognizable. Keep ambiguous program submissions in an ownership hold with an explicit recovery choice. Echo controls presentation; it does not authorize sending.

• Ctrl+S sends the capture once to the same live command. Tab/Ctrl+Q reviews without replaying Enter. Later likely/visible reports never approve the capture.
• Completion invalidates delivery and keeps hidden prompt review; cancellation, terminal handover and shutdown clear pending captures. Existing approval entries never auto-send.
• Parse real editor keys and NMSh actions before guarding replayed program Enter; preserve queue-review privacy across later hidden-reader reports. No detector, protocol, dependency or CI-gate change.

Refs #473. Separate from test-only PR #513, whose exact head and failed comprehensive gate remain untouched. No automatic merge or release.

Validation: full npm run verify passed locally with the intended truecolor profile and shortest compliant T7 test root. Fixed two-round/two-worker real-PTY stress passed with no lifecycle leftovers. Deterministic coverage includes unknown/older lookups, late likely/hidden inference, completion, cancellation, native handover, editor actions and ordering. Exact-head hosted CI: normal PR run40630119387 PASSED; single comprehensive run40630147755 FAILED macOS1/3 in browserConsoleChrome.test.ts:106. Its Mermaid-render result was neither success nor syntax; the actual code/reason was not logged, so causation remains unconfirmed. Other test/platform/timing/Fedora gates passed; aggregate checks correctly failed. No retry. This PR stays draft/unmerged until that independent blocker is understood and the required gate is satisfied. Controlled old/current-dev source fails both immediate/delayed approval-enqueue assertions; corrected source retains the real read and recovers exactly once after explicit Ctrl+S. Independent review findings are addressed. Physical Ghostty/Zed QA remains NOT RUN.

Manual validation on Ghostty and Zed: run a prompt-less zsh read and type an immediate answer; when held, Ctrl+S should deliver it once without queueing. Check hidden reads show no answer; sleep plus Enter remains held; Ctrl+Q reviews for queueing; Tab does not submit; Esc/Ctrl+C cancel; completion keeps review-only typing; Fish/native and fullscreen handover receive only their own new input.

Do not integrate #513 yet: preserve its exact head and failed comprehensive evidence. After this gate is understood/corrected, use a separately pinned integration candidate containing current dev, the reviewed correction and exact #513 head, then run the unchanged comprehensive gate once. A normal green rerun does not clear either failure.

Deferred resource investigation plan preserves all prior performance benefits, CPU/RSS costs and measurement limits; no memory optimization was started.

[raiseCatError]

Cross-reference from #541. I didn't change this branch.

• The stuck "Can't tell whether echo is reading your typing" state. It came from zsh history expansion, not an ownership race. In echo "STEP 3 — Finished!", !" removes the closing quote, so zsh waits at an invisible PS2. No exec ever arrives, the look returns unknown, and a bare Ctrl+C at PS2 does nothing in zsh without ZLE. Queue composer, unfinished command lines and owner-side reply hold (stacked on #514) #541 detects the extra continuation prompt, cancels the line, explains it and gives it back.
• "Awaiting approval" during sleep. This is the Under heavy load, a reply typed right after a prompt-less read starts can be queued instead of answering it #426 quarantine working as designed. Queue composer, unfinished command lines and owner-side reply hold (stacked on #514) #541 adds opt-in Queue Mode for people who want Enter to mean "queue".

This PR makes Enter during a command that isn't reading a hidden confirmation hold, which pulls the opposite way from Queue Mode. Before either merges, we should decide whether this PR is still needed with #541, and if so, how the two compose. I'd suggest rebasing it onto #541 and keeping the hold only when Queue Mode is off.