refactor: schema region consolidation + playback cap + deep-agent fold

[EtienneLescot]

refactor: schema region consolidation + playback cap + deep-agent fold

Pure refactor that bundles three related shrinkages from the v1.8.0 ponytail
audit. No behavior change visible to the user; the public types and the
schema-version contract are byte-identical.

Changes

1. src/lib/ai-edition/schema/index.ts — region-schema refine helper

   Five near-identical region schemas (rangeSchema, trimRangeSchema,
   gapSchema, annotationRegionSchema, zoomRegionSchema) all carried
   the same .refine((data) => data.end >= data.start) clause with
   field-name-keyed message + path. Extracted a small generic endGteStart
   helper that takes the schema and the two field keys, and threaded the
   five call sites through it. The parse output (including the error
   message + path) is byte-identical to the pre-refactor wire format.

2. src/components/video-editor/types.ts — MAX_PLAYBACK_SPEED = 16

   The 100× editor cap is unreachable in the HTMLMediaElement path:
   Chromium hard-caps playbackRate at 16 (the comment at the constant
   already documents this). The 16–100 range is the export-time stretch
   path, and the same clampPlaybackSpeed was used for both preview and
   export, so the input parser allowed values the preview can't render.
   Dropped the cap to MAX_NATIVE_PLAYBACK_RATE = 16, and the existing
   clampPlaybackSpeed now rounds + clamps to that tighter ceiling. A
   50× entry now clamps to 16 instead of 100.

   SpeedControl.test.tsx and customPlaybackSpeed.test.ts had
   boundary cases that pinned the 100 ceiling — updated mechanically
   to pin 16 (the new max, with a "16.1 is too-fast" case to assert the
   new behavior, replacing the "16.1 is valid" case the old 100 cap
   allowed).

3. electron/ai-edition/deep-agent/ — fold capability table into chat-model.ts

   agent-provider-capabilities.ts (capability tables, reasoning-effort
   normalization, thinking-block wiring for OpenAI/Anthropic/MiniMax/
   OpenRouter/Google) was only consumed by chat-model.ts. Folded the
   file's contents into the top of chat-model.ts as a "per-provider
   reasoning-effort capability table" section, and deleted the standalone
   file. The service.ts (deep-agent runtime) and its chat-service.ts
   consumer are unchanged — they own a different concern (the LangGraph
   thread + tool graph), not the chat-model factory. The capability
   tests at agent-provider-capabilities.test.ts keep their path and
   were re-pointed at ./chat-model; the suite still passes 17/17.

Diff stat

 8 files changed, 332 insertions(+), 317 deletions(-)
 delete mode 100644 electron/ai-edition/deep-agent/agent-provider-capabilities.ts

Type

• Refactor / maintenance
• Bug fix
• New feature
• Documentation

Severity

• Major
• Minor
• Patch

Platform

• Not platform-specific
• macOS only
• Windows only
• Linux only

Testing

• npx tsc --noEmit — clean exit 0.
• npm run test — 1144 / 1144 pass, identical to the v1.8.0
  baseline.
• npm run lint — 0 errors, 0 new warnings (7 pre-existing warnings in
  NewProjectModal.test.tsx, rafCoalesce.test.ts,
  useNativeCompositorView.ts; all in files untouched by this PR).
• src/lib/ai-edition/schema/index.test.ts — 31/31 pass without
  modifying test expectations; the parse output is byte-identical for
  representative inputs (rangeSchema, trimRangeSchema, gapSchema,
  annotationRegionSchema, zoomRegionSchema).
• electron/ai-edition/deep-agent/ — 17/17 pass after the fold
  (capability tests + chat-model tests).

No manual smoke test required: changes 1 and 3 are pure refactors with
byte-identical parse output; change 2 is a constant tighten that
only affects the input-parser ceiling (the 16–100× export path is
unchanged, the input parser now rejects what the preview can't render).

Notes for reviewer

• MAX_PLAYBACK_SPEED and MAX_NATIVE_PLAYBACK_RATE are kept as
  separate exported names with the same value. The renderer
  (VirtualPreview.tsx) reads MAX_NATIVE_PLAYBACK_RATE and the
  parser/persistence read MAX_PLAYBACK_SPEED; merging them is a
  rename, not a refactor, and was left for a follow-up.
• The endGteStart helper is a single-call-site style: each region
  schema calls it once with its own endKey/startKey. The 3-line
  factory shape suggested in the audit was rejected because gapSchema
  uses timelineStartSec/timelineEndSec (not startSec/endSec),
  so a 'Ms' | 'Sec' key factory wouldn't fit cleanly.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 71b31df2-8a49-4834-93fb-e498b5c92f1b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch ponytail/schema-and-module-cleanup

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}