feat(desktop): replicate Codex computer-use cursor - #1255
Conversation
|
CI follow-up: the failure was the latest main branch's repository-wide format baseline in |
Astro-Han
left a comment
There was a problem hiding this comment.
Two P2 comments on cursor edge behavior. Neither rises to P1, so I am approving with comments.
|
@Astro-Han Both P2 comments are addressed in 5395fb2: viewport-aware path scoring/direct fallback plus cancel-opacity reset, with regressions for the reported edge path and cancel-during-fade. The refreshed typecheck, test, and e2e checks are all green. Re-review when convenient. |
|
LGTM now! Merging. |
Summary
Reverse-engineering boundary
The glyph path, center hotspot, thresholds, and all recovered motion constants come from the signed 2026-07-16 SkyComputerUseService binary. The native candidate scoring function is stripped, so the planner is a structural reproduction using the exact recovered handle and arc constants; it is not presented as instruction-for-instruction equivalence.
Base CI repair
The latest main branch failed the repository-wide format check in
packages/storage/src/__tests__/settings-store-usage.test.ts. A separate mechanicalstyle(storage)commit applies Biome's formatting only, with no semantic changes, so this PR can run against a clean baseline.Visual validation
Validation