[kantord]

Summary

The Escape double-press check compared with <= 500 while the expiry timer fires at 500, so the exact boundary raced. One rule now: less than 500 ms.

Part of #2303 ("Tiny follow-up fixes"); it does not close it. This is PR 8 of 14 in a stack of draft PRs, reviewed and merged in order.

What changed

• chat-escape.ts: <= 500 becomes < 500; the redundant changed alias is removed (the focusin listener stays, so a focus move still cancels an armed press).
• Tests: the 499 ms and 500 ms boundary (the clock is moved without firing the expiry timer, which would hide the race); held Escape (repeat: true) at each layer (ask deny, plan iterate, panel close, stop); a same-tick double-click on an ordinary verdict button on the chat and on the side thread.

Behavior changes

A second Escape exactly 500 ms after the first no longer clears the draft.

Verification

Biome, typecheck, the full Studio test suite and pnpm generated:check pass on this branch, run on the assembled stack. The full Playwright suite passes on the stack tip (122 passed, 30 skipped, 0 failed, including chat-approval against a real mecated build). Each new test fails with its fix reverted. The existing "one verdict" test only proved the disabled state (fireEvent flushes between clicks); the new test uses a real same-tick double-click. The plan version already existed and was not duplicated.

Not included

A planted-secret-in-ask-args test: scrubbing happens in the engine and the UI renders args as given.

Fully or partially written by an AI agent.

🤖 Generated with Claude Code

[peppescg]

Thanks Daniel. The < 500 rule now agrees with the expiry timer, and moving the clock without firing the timer is the right way to pin the boundary.

Nits:

• apps/web/src/features/chat/chat-escape.ts:124: the re-arm branch replaces expiry.current without clearing the pending timeout. A press at exactly 500 ms (or later, if the timer is delayed) now re-arms while the first timer is still queued. That timer then fires and immediately disarms the new arm, so the hint flashes and the next press re-arms instead of clearing. Add if (expiry.current !== undefined) window.clearTimeout(expiry.current); before scheduling the new timeout. A test could advance timers by a few ms after the 500 ms press and assert the hint is still shown.
• apps/web/src/features/chat/chat-escape.test.tsx:292-295: the ordinary-ask block only checks that a held press does nothing. Follow it with a normal press and expect "deny", as the plan block does with "iterate", so the test can't pass on a layer that never acts.