[moikdev]

Stacked on #165 (base fix/batch2-anki). The chain is #163 → #164 → #165 → this PR. They all touch the subtitle/translation model wiring (src/models/subs/init.ts), so the independent branches conflicted. Merge in that order.

Playback, input and timeline fixes from the 30 Sep / 2 Oct audits.

Fixes

Input

• Fixes [P2] Disabled extension still intercepts keyboard and mouse playback controls #111: a disabled extension still intercepted keys and mouse buttons. Keyboard shortcuts and mouse bindings now do nothing while Himotoki Sub is disabled. Disabling it also stops any replay, slow replay or loop it started.
• Fixes [P3] Holding a toggle shortcut repeatedly opens and closes panels #127: holding a toggle key flickered panels. Toggle shortcuts (D, B, T, R, L, \, H) now act once per press and ignore OS auto-repeat. The speed steps (, and .) still repeat.
• Fixes [P2] Subtitle hotkeys intercept typing in plaintext-only contenteditable editors #128: hotkeys fired while typing in plaintext-only editors. The editable check now uses isContentEditable and the contenteditable inheritance rules (including plaintext-only, and an explicit false stops it). It also follows the composed path into shadow roots.
• Fixes [P3] Assigned middle-click playback action also cancels links over the video #137: middle-click on a link over the video was swallowed. An assigned mouse action now leaves links, dialogs and menus over the player to the browser.

Replay / auto-pause

• Fixes [P2] Old slow-replay timers override newer speed choices and cut subsequent replay short #108: old slow-replay timers overrode newer speed choices. Slow replay now ends in media time and restores the current speed setting. A newer slow replay, a speed change, a new video or disabling the extension cancels it.
• Fixes [P2] Seeking forward while replaying a subtitle pauses the newly selected scene #129: seeking forward during a replay paused the new scene. Replay pauses only when playback runs through the cue end. A seek forward past the end, or backward, hands the video back without pausing. Any explicit seek (transcript, timeline, history) cancels the replay.
• Fixes [P2] Replacing captions leaves shadowing locked to the old subtitle’s loop window #130: the shadowing loop stayed on the old captions' window. When captions change, an active loop follows the same line (same id and text, e.g. after a delay resync) or ends.
• Fixes [P2] Auto-pause misses cue endings when time updates skip its narrow pre-end window #131: auto-pause missed cue ends. It now also pauses when sparse timeupdates skip the 250 ms window, as long as normal playback crossed the end. Seeks never pause, and resuming doesn't pause at the same line again.
• Fixes [P2] Leaving subtitles resumes playback after a newer manual pause #142: leaving the subtitles resumed a newer manual pause. The hover pause is dropped once playback starts by any other means, so leaving the line keeps a later manual pause. It is also reset when the video element changes.
• Fixes [P3] Adjacent subtitle cues both remain active at their shared boundary #121: two adjacent cues were both active at their shared boundary. Active cues are now half-open [start, end) intervals for both the primary and secondary tracks, so only the next cue is active at a shared boundary. Zero-length cues are never active; real overlaps still show together.
• Fixes [P2] Replacing the video leaves the old timeupdate listener attached #155: the old video kept its timeupdate listener. Exactly one listener now lives on the current video. A replaced element is detached and can no longer drive the clock.

Timeline

• Fixes [P2] Paused videos keep rescanning the entire subtitle track and publishing React state every animation frame #151: paused videos kept rescanning the track every frame. The timeline indexes the track once and finds visible cues with a binary search. A frame with unchanged time and width (a paused video) does no work and publishes no state.
• Fixes [P3] Subtitle timeline hides cues that span its entire visible time window #152: cues spanning the whole window were hidden. Visibility is now interval overlap (end > from && start < to), so a spanning cue shows. Markers are clipped to the window.
• Fixes [P2] Clicking a subtitle timeline marker seeks relative to the marker instead of the timeline #150: clicking a marker seeked relative to the marker. A click now maps clientX through the timeline's own rectangle (clamped), not the clicked element's offsetX.

Visible changes

Only the subtitle timeline changes: a cue spanning the whole 30 s window now draws a full-width marker, and markers no longer extend past the bar's edges. I could not capture a screenshot, because the in-player timeline isn't rendered by the repo's fixture harness and needs a live player.

Verification

• pnpm exec tsc --noEmit: clean. pnpm lint: 0 errors.
• pnpm test:unit (full stack): 48 files and 430 tests pass. New tests:
  • inputTargets.test.ts, inputHandlers.test.ts: real handlers with DOM doubles.
  • slowReplay.test.ts, replayVideoClip.test.ts: fake media clock.
  • cuePlayback.test.ts, playbackWiring.test.ts: the real video/subtitle Effector graph.
  • cueWindow.test.ts: randomized comparison against a full scan, overlap policy, click geometry.
  • getCurrentSubs.test.ts, videoClock.test.ts.
• pnpm build + pnpm test (Chromium regressions): all PASS, 0 runtime errors. A new browser check covers [P2] Leaving subtitles resumes playback after a newer manual pause #142: hover pauses, leaving resumes, and a manual play+pause made over the line survives leaving it. I confirmed that this check fails without the fix.
• I ran the audit evidence tests for [P2] Old slow-replay timers override newer speed choices and cut subsequent replay short #108 and [P2] Disabled extension still intercepts keyboard and mouse playback controls #111 (slow-replay, disabled-input, which assert the buggy behaviour). The disabled-input cases no longer reproduce. The slow-replay doubles have no addEventListener and can't drive the new media-time code; slowReplay.test.ts and playbackWiring.test.ts cover those scenarios instead.

🤖 Generated with Claude Code

[moikdev]

Hermes review failed

Hermes could not produce a code-review verdict for this head SHA.

Engine: agy/gemini-3.8-flash-high
Review mode: full
Head: de657a260b7eff538e356ec150acc425d1262d88
Generated: 2026-10-04T11:31:22+00:00
Failure phase: engine

Safe summary:

agy failed (3)
STDOUT:
STDERR:
error: Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 91h32m44s.
AGY_ERROR: {"short_error":"RESOURCE_EXHAUSTED (code 429): Individual quota reached. Please upgrade your subscription to increase your limits. Resets in 91h32m44s.","status":"RESOURCE_EXHAUSTED","error_code":429,"code_kind":"http","retryable":true,"error_id":"c5e77849-fbff-4d65-ba98-eda2dfdea9cf-1"}

Local logs: /home/moik/.hermes/logs/review-bot/himotoki-sub.log

[moikdev]

/hermes review