Closed in error#2050
Closed
devtronic-me wants to merge 95 commits into
Closed
Conversation
…ions Slice 0 (Discovery) and the first piece of foundation work for Slice 1: - docs/: 16 discovery docs across product, architecture, hub/node protocol, plugin lifecycle & extension points, data model, auth/RBAC, device management, session lifecycle, UI inventory, devops, testing, observability, security, performance, plus strengths/weaknesses/risks synthesis and 38 open questions. Tick Slice 0; replace the speculative Slice 2+ list in docs/SLICES.md with a concrete prioritized list driven by Discovery. Slice 1 spec drafted at docs/slices/01-falx-baseline.md (kept as reference; Slice 1 will be reshaped before execution). Execution plan record at docs/superpowers/plans/. - falx-ui/: one-time copy from dashboard-frontend/. dashboard-frontend/ stays untouched per CLAUDE.md hard rule. - buildAndCopyWeb.sh: switched to building falx-ui/ instead of dashboard-frontend/. Plugin still serves from src/public/. - falx-ui/vite.config.js: dev server proxies /device-farm to localhost:4723 so `npm run dev` HMR runs against a live appium plugin backend without rebuilds. - CLAUDE.md: project context, naming rules, upstream-relationship workflow, hard rules, slice methodology. - .gitignore: ignore .playwright-mcp/ test artifacts. - package.json: pin sharp@0.34.4; add appium-device-farm@^11.3.2 dev dep. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace the ProtectedRouteWrapper body — swaps the bg-gray-900 div + <Header /> with <AppLayout> — and delete Header.tsx / NewHeader.tsx which are now dead code. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move Falx logo + wordmark from sidebar header into the left edge of TopBar. Sidebar now starts at top:56 (z-40) so the top bar spans full viewport width unbroken. AppLayout renders TopBar first then Sidebar+main side-by-side. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sidebar now starts at top-0 with z-50 (covers the top-left corner of the top bar), carrying the Falx brand header in an h-14 block. TopBar drops to z-40, removes the brand, and uses dynamic paddingLeft from pinned state so the page title aligns with content below the sidebar edge. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…dings Spike 01 (Android via Tango): GREEN — Tango drives upstream scrcpy-server-v3.3.3 with first frame in 539 ms, 5-min run stable, taps sub-pixel accurate. Cleared the way for the Falx browser device-use slice on Android. Spike 02 (iOS via WDA + go-ios), Phase 1: - Streaming PASSES — 19.08 fps over 5 min, max gap 187 ms at tuned WDA settings (framerate 20, quality 70, scaling 100). No Phase 2 (WebRTC) needed for streaming quality. - Tap injection FAILS on iOS 26.4.2 with WDA 12.2.2: every coordinate and element-based tap endpoint returns 200 null and no-ops on the device. Reads / element finders work; writes don't. Needs a separate follow-up spike to enumerate working tap channels before iOS device-use can ship. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ssons Spike 02 found that WDA 12.2.2 on iOS 26.4.2 silently no-ops all tap APIs (coord, W3C, element-based) while reads keep working. That breaks the iOS device-use feature and needs its own follow-up before any slice. - docs/spikes/03-ios-tap-injection-spike.md — plan to enumerate candidate tap channels (extended WDA caps, /appium/settings tunables, latest WDA rebuild, Sauce/HeadSpin forks, go-ios direct HID, tidevice). Run cheapest first, stop at first channel that passes tap + swipe + typing on iOS 26. - docs/spikes/README.md — list spike 03 as the next required spike. - docs/BACKLOG.md — log the WDA + iOS 26 regression, the runwda bundle-ID convention shift for iOS 17+, the RSD parallel-spawn race, the dropped /wda/tap/0 endpoint, WDA's non-standard MJPEG boundary, the single-client MJPEG constraint, the unlock-bypass limitation, and the pixel-vs-points tap-coordinate gotcha. - .gitignore — ignore selfIdentity.plist (go-ios's device-pairing identity, contains a private key). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
First slice replacing the proprietary appium-device-farm@11.3.2 "Use Device" feature with a Falx-owned, MIT/Apache-licensed implementation on top of Tango (yume-chan) + scrcpy. Vertical-per-platform; iOS lands in a follow-up slice. Plan is subagent-driven with two-stage review. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…cpy-server.jar
ADR 0001 documents the Tango + WDA decision. Pins:
- server: @yume-chan/{adb,adb-server-node-tcp,adb-scrcpy,scrcpy,stream-extra}
- falx-ui: @yume-chan/{scrcpy,scrcpy-decoder-webcodecs,stream-extra}
- on-device: scrcpy-server-v3.3.3.jar (sha256 7e70323b...)
Refs: docs/superpowers/specs/2026-05-16-android-use-device-design.md
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ight (TDD) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…maxBuffer Plain 'ios apps' emits full plist metadata (~3.4 MB on a typical device) which blows Node's default 1 MB execFile maxBuffer; the catch then miscategorises the buffer overflow as WDANotInstalledError. '--list' returns bundle-id-only lines (~8 KB) and still contains the runner.
…unner go-ios v1.0.188+ enforces 'all-or-none' on the bundleid/testrunnerbundleid/xctestconfig trio in 'ios runwda'. Spike 02 used an older version that accepted partial flags. With no flags passed, go-ios auto-discovers the single installed XCTest runner — more robust to varying signed bundle names and xctest target names than hardcoding xctestconfig=WebDriverAgentRunner.xctest.
…onfig go-ios v1.0.188's 'no flags' path falls back to hardcoded Facebook defaults (com.facebook.WebDriverAgentRunner.xctrunner) rather than scanning installed apps, so any non-Facebook re-signing fails to launch. Pass bundleid + testrunnerbundleid (the installed runner) plus xctestconfig=WebDriverAgentRunner.xctest (canonical WDA test-target name, unchanged by re-signing).
aspect-ratio on a canvas element is unreliable: its intrinsic backing- store dimensions (width/height HTML attrs) override CSS sizing when both CSS dimensions are auto. Computed style would resolve to e.g. 1352×700 instead of the locked 323.5×700 portrait shape — the image rendered into the misshaped canvas appeared stretched horizontally. Fix: wrap canvas in a regular block element that carries the aspect- ratio + definite height; have the canvas fill it via width/height: 100%. alignSelf: center on the wrapper opts out of cross-axis stretch in the flex-column parent so the wrapper shrink-wraps to height × aspect. Diagnosed via Playwright DOM inspection on the live session.
…er WDA
Appium's xcuitest driver always POSTs to WDA's /session when attaching
via webDriverAgentUrl, displacing our bridge's session. Every subsequent
/session/<stale>/wda/{tap,drag,activeAppInfo,apps/terminate} call from
the WS handler then 404s — taps, swipes, Home and App Switcher all dead.
After Appium createSession succeeds, probe any unknown WDA path; WDA
includes the current sessionId in every JSON response. Rebind
bridgeHandle.sessionId, then re-apply MJPEG settings (they're session-
scoped on some WDA builds and would otherwise revert to defaults).
Verified end-to-end on kry-phone with Playwright: tap, Home, App
Switcher all changed the canvas hash and the backend log shows the
rebind line with no dispatch 404s.
Records Playwright-driven live results, the five fix commits, open validation items, and the iOS vs Android architecture limitations. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sequential start→stop loop records p50/p95/max timing and lets the operator confirm no ios runwda / forward process leaks post-cycle. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Task 13 Steps 1-7 verified via Playwright. Steps 8-13 marked pending operator validation. Tasks 14/15 marked pending second iPhone / Safari+Firefox. Task 16 Step 1 ticked (script written). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
50 sequential start->stop cycles against kry-phone: - Start: p50 5331ms / p95 5773ms / max 5788ms (under 8s budget; latency trended *down* over the run — warm cache, no leak). - Stop: p50 26ms / p95 31ms / max 59ms (HTTP ACK only; teardown runs async via registry stop closure). - Zero leaked 'ios runwda' or 'ios forward' processes post-run. - Port allocator reused slot 0 (8100/9100) cleanly across all cycles.
Captures the two operator-flagged gaps on the iOS slice for a fresh
session to pick up:
A. Lock-screen probe doesn't trip the 423 path on iOS 26.
B. Tap/swipe responsiveness is well below Android — root cause is
WDA's per-input HTTP round trip + MJPEG observe lag. Lists four
candidate levers (optimistic UI, W3C actions, WebRTC, drop Appium
attach) without picking one.
Includes a kickoff prompt to start the new session against this doc.
Brainstorm output for the iOS input-responsiveness follow-up. Spec covers live drag tracking, optimistic canvas overlays, and a duration-cap fallback to bring iOS Use Device input feel to BrowserStack parity. Plan is 19 tasks across spike (Phase 1), implementation (Phase 2 with two conditional branches), and verification (Phase 3 including agent-side Playwright MCP self-verification). Picks up from feat/ios-use-device. Issue A (lock probe) deferred to a separate slice; WebRTC deferred; Appium attach kept (gestures already bypass it). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase-1 spike (docs/spikes/04-ios-input-latency-spike.md) measured WDA dispatch wall-time on iOS 26.4.2 at 600-1800ms per call — 20x worse than the original spec's working assumption (30-80ms). Keep-alive doesn't help; only 2 of 8 attempted runtime tunables stick. True live drag tracking is therefore not achievable in this slice with vanilla WDA. Operator decision: ship the achievable wins now. Spec amended with a "Revision" section at top noting the scope cut. Plan rewritten from 19 tasks (live drag, pointer state machine, streaming pointer wire protocol) to 7 focused tasks: 1. Apply WDA tunables on session start (~40% dispatch reduction) 2. Add WDAClient.actions() + tapViaActions() helpers 3. Swap tap dispatch to /actions + cap swipe duration at 120ms 4. Optimistic canvas overlays (ring + trail) for falx-ui 5. Run full unit-test suite 6. Agent Playwright MCP self-verification against running app 7. Operator subjective gate on kry-phone Acceptance bar reframed from "BrowserStack-grade" to "noticeably more responsive Y/N". Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reduces per-call WDA dispatch wall-time by ~40% on iOS 26.4.2 by setting waitForIdleTimeout=0 and animationCoolOffTimeout=0 immediately after WDA session creation. Other tunables attempted in the spike were not recognized by WDA 12.2.2 and are documented for future revisits. See docs/spikes/04-ios-input-latency-spike.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…helpers Spike 04 found /actions p50 ~631ms vs /wda/tap ~818ms with tunables applied. Adds the W3C Actions wrapper and a 1-frame tap helper for the router to swap to in the next commit. See docs/spikes/04-ios-input-latency-spike.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… at 120ms Two iOS input-responsiveness changes: - Tap dispatch now uses WDA's W3C /actions endpoint via tapViaActions (p50 ~631ms vs /wda/tap ~818ms after tunables, per spike 04). - Swipe duration is clamped to [50, 120]ms regardless of how long the operator drew the gesture. Eliminates the 'WDA replays the swipe slowly on the iPhone' complaint without breaking tap/swipe distinguishability. See docs/spikes/04-ios-input-latency-spike.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…trail) Draws a transient ring at pointerdown and a fading trail during drag, layered over the MJPEG frame on the same canvas. Renders instantly, no network round-trip. Masks the unavoidable ~600ms WDA dispatch wait on iOS 26.4.2 (see docs/spikes/04-ios-input-latency-spike.md) by giving the operator immediate visual confirmation that the gesture was registered. Pure-function renderer with unit tests for ring drawing, trail polyline, expiry, and trim semantics. Also installs vitest + wires test script in falx-ui (was absent; required for the spec file). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Previously the MJPEG handler called drawImage which clears the canvas bitmap, erasing the optimistic ring + drag trail overlays for up to ~16ms (one rAF tick) at every MJPEG frame arrival. On a static overlay that's a visible 20Hz flicker. Fix: invoke renderOverlays synchronously after drawImage so the overlay is restored same-tick. The rAF loop continues to drive the fade-out animation between MJPEG frames. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Retroactively ratifies the vitest dep addition that landed as part of commit 9cab3d7 (iOS canvas overlay renderer + unit tests). vitest is the right test runner for falx-ui (Vite + React, zero-config); future falx-ui slices use it freely. Server-side stays on mocha + chai (no change to existing convention). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
After the iOS input responsiveness slice landed and a research subagent confirmed vanilla WDA is the universal OSS ceiling (GADS, Sonic, ControlFloorAgent, STF-iOS, atxserver2, tidevice all use WDA), document: - The 600-1800ms per-call WDA latency floor as a "lesson" so future iOS work doesn't re-discover. - The Appium issue #16230 WDA-patch spike as a deferred tech item — one identified local lever, ≤2 days, would patch the accessibility- snapshot calls out of WDA's tap/actions hot path. - tddworks/baguette tracking for real-device iOS HID injection. See docs/spikes/04-ios-input-latency-spike.md for the data. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Agent Playwright MCP self-verification (2026-05-19): dashboard loads, Use Device → stream renders, WS protocol correct (CLIENT_TAP_TAG 0x20 / 9 bytes), end-to-end pipeline works (synthetic tap on Photos icon at iOS-points coords launched Photos on the device), optimistic ring overlay paints before WS round-trip (pixel sample RGB 234,231,226 at expected ring-stroke location), clean teardown (zero leaked ios runwda / ios forward processes). - Operator manual gate (2026-05-18): "noticeably more responsive Y, but still way behind Android" — accepts shipping the wins given the WDA ceiling. Diagnostic confirmed: streaming half is real-time; dispatch half is the bottleneck. - Slice outcome: at the OSS state-of-the-art ceiling for iOS 26.4.2. WDA-patch spike (Appium #16230) logged in BACKLOG for future round. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Author
|
Accidentally targeted upstream; re-opening against my own fork. Apologies for the noise. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Accidentally targeted upstream; closed.