Screenpipe upstream contribution burst 2026-07-03

decision Fri Jul 03 2026 00:00:00 GMT+0000 (Coordinated Universal Time) screenpipeupstreamcontributionreview

Screenpipe upstream burst — 2026-07-03

Session summary

Two grill-me waves ran comprehensive audits against screenpipe/screenpipe (fast-moving Rust+Tauri monorepo, 2000+ upstream stars). Fork stayed byte-clean per no-fork-divergence. All 37 contributions filed atomically from chirag127/<branch> heads against upstream/main.

Wave 1 — reactive (user hit onboarding stuck)

Root cause: SAP CPIT DLP intercepts WebView2 → localhost:3030 from the same Tauri process (curl works from outside, webview does not).

7 PRs:

#4851 401 detection + actionable copy on live-feed polls
#4852 SCREENPIPE_SKIP_ONBOARDING=1 env-var escape hatch
#4853 sdk.yml concurrency group
#4854 earlier skip button in onboarding (10s + stuck state)
#4855 AGENTS.md H1 typo fix
#4856 crossbeam → dev-dependencies
#4857 tsconfig noFallthroughCasesInSwitch

4 issues:

#4850 corp-DLP WebView→localhost interception (reproducer)
#4858 3 timer/interval leaks in Tauri hooks
#4859 tsconfig strictness baseline drift across 6 configs
#4860 10 more workflows missing concurrency groups

Wave 2 — proactive comprehensive review

Six parallel subagents (UX/perf/security/architecture + deep-research + live-metrics extraction from user's own running engine).

Deep-research findings (4 axes, sourced 2026-07): Tauri v2 patterns, screen-recording competitors 2026, MCP spec 2026-07-28 RC (breaking), Rust pipeline patterns.

Live metrics from user's 3h session: OCR cache 0% hits, 33% empty transcriptions, 1.5GB app RAM, 100%+ single-core CPU sustained.

12 more PRs:

#4866 permission-buttons a11y (icon-only aria-label)
#4867 settings-search empty-state copy
#4868 rewind-search a11y (thumbnail loading + errors)
#4869 engine-startup streaming region aria-live
#4873 audio-dedup: drop 2 String allocs per chunk (24/7 hot path)
#4875 ai-gateway README (7 providers documented)
#4876 audio-chunk deletes: 2N stmts → 2 stmts (IN clause batching)
#4879 MCP README: document the 23 tools that were missing
#4881 analytics.rs: respect analytics_enabled during attribution beacon
#4884 README FAQ: correct "no cloud data" claim (PostHog+Sentry ship enabled)
#4887 hardened runtime on beta+enterprise macOS builds

14 more issues:

#4862 Tauri capabilities/main.json hardening (auto-migrated + wildcards)
#4863 MCP spec 2026-07-28 RC tracking + Resources migration
#4864 tokio-console feature-gate for task-starvation observability
#4865 3h session perf data (OCR cache miss, empty transcripts, RAM)
#4870 onboarding stream error copy on corp networks
#4871 "watching · Ns" indicator a11y strategy
#4872 20+ icon-only buttons need aria-label sweep
#4877 FPS derate on battery/thermal, not just idle
#4878 backfill_meeting_segment_speakers query pattern
#4880 background poll intervals configurable + DeviceTier-adaptive
#4882 ai-gateway missing Groq + Ollama providers
#4883 screenpipe-engine 12-sibling coupling / trait extraction
#4885 PostHog+Sentry init before settings hydration
#4886 SecretStore base64-fallback when encryption off

Discipline maintained

Every contribution:

Branched off upstream/main (never fork main)
Read the file before editing (never grep-only)
≤150 word body per terse-issues-less-hallucination
Thanks-maintainers on last line
Preserved screenpipe file headers on every touched .rs/.ts/.tsx
≤5 files per PR (most = 1 file)

Parallel-agent turbulence

7+ concurrent subagents on same submodule caused branch clobbering / stash races. Recovered every commit via git reflog + push-by-SHA. Never used git reset --hard on fork state. Lesson: --isolation worktree for parallel fork work. One subagent (architecture review) used worktree correctly at repos/frk/screenpipe-review-arch/ — cleaned up post-run.

Cross-refs

unstuck-screenpipe-onboarding-corp-vdi — the reactive-session runbook
no-fork-divergence — how we stay clean
terse-issues-less-hallucination — body constraint

Follow-up if any land

If maintainer merges > 60% of PRs, this session becomes a case study for the fork-thin-upstream-tracking + audit-loop pattern. Track merge rate over next 30 days.