Summary

• Wrap startup capture in the same 15s timeout used by live captures so a stuck first Windows/UIA capture cannot block the monitor loop forever.
• Gate visual-change screenshots while keyboard activity is still recent, avoiding redundant background frame-diff captures during typing-driven full captures.
• Keep visual checks active after keyboard activity goes quiet, so mouse-only movement does not suppress passive visual capture.

Verification

• cargo fmt --package screenpipe-engine
• cargo build -p screenpipe-engine --profile release-dev --no-default-features
• 5-minute Windows CLI benchmark with simulated Notepad typing, audio/telemetry/meeting detector disabled:
  • before startup timeout fix: UI events persisted but 0 frames
  • after startup timeout fix: 77 frames / 77 UI events
  • after keyboard-only visual gate: mean CPU 3.579%, median 2.972%, p90 7.411%, max 12.394%, with 77 frames / 77 UI events / 77 non-empty accessibility frames

Notes

• Focused cargo test did not complete in this local workspace: debug target filled the system drive, and release-dev test harness compilation timed out. The release-dev build and runtime benchmark completed successfully.