Record video on every CI trail run, and keep recordings in sync on older ffmpeg #773

Workflow file for this run

.github/workflows/pr-checks.yml at d1d301d

	name: Checks

	on:
	# Runs on PRs targeting the default branch
	pull_request:
	branches: [ "main" ]

	# Runs after merging to the default branch
	push:
	branches: [ "main" ]

	# Allow only one concurrent deployment, skipping runs queued between the run in-progress and latest queued.
	# However, do NOT cancel in-progress runs as we want to allow these production deployments to complete.
	concurrency:
	group: ci-${{ github.ref }}
	cancel-in-progress: true

	jobs:
	# Pre-build the trailblaze uber JAR once, with the WASM report template
	# embedded, and publish it as an artifact for the downstream test jobs to
	# consume. Mirrors the trailblaze-internal `build-uber-jar` Buildkite step:
	# downstream test jobs install the prebuilt JAR via
	# `scripts/install-trailblaze-from-artifact.sh` instead of paying their own
	# `./gradlew :trailblaze-desktop:jar` (~3–5 min cold) plus a cold daemon
	# start via `./gradlew :trailblaze-desktop:run`. The installed `trailblaze`
	# binary runs the JAR directly via `java -jar`, skipping Gradle entirely
	# for every CLI invocation.
	#
	# `-Ptrailblaze.wasm=true` (via TRAILBLAZE_GRADLE_EXTRA_ARGS) bundles the
	# WASM-rendered report template into the JAR's classpath, so every
	# downstream `trailblaze report` invocation finds its template inside the
	# uber JAR — no separate template download dance is required.
	# (The internal repo keeps the CI JAR wasm-less and downloads the template
	# as a sibling artifact; OSS keeps it simple and embeds.)
	build-uber-jar:
	runs-on: ubuntu-latest
	timeout-minutes: 20

	steps:
	- name: Checkout
	uses: actions/checkout@v4

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Gradle
	# Persists the Gradle User Home (dependency cache + local build cache)
	# across runs. Pushes to main seed the cache; PR runs restore it
	# read-only, so warm PR builds pull unchanged task outputs — including
	# the ~5-minute Kotlin/WASM compile+optimize chain — from cache.
	uses: gradle/actions/setup-gradle@9c971963bec38e04b3d30dcc455b5382be2fdbfb # v6.3.0
	with:
	cache-read-only: ${{ github.ref != 'refs/heads/main' }}

	- name: Setup Android SDK
	# trailblaze-desktop transitively depends on Android-targeted modules,
	# so the Android Gradle Plugin needs an SDK to evaluate the build.
	uses: android-actions/setup-android@9fc6c4e9069bf8d3d10b2204b1fb8f6ef7065407
	with:
	# Drop the action's default `tools` package. Google removed the obsolete SDK Tools
	# package from the repository, so `sdkmanager tools` fails outright on runner images
	# that no longer ship it preinstalled. `platform-tools` is what the build needs.
	packages: platform-tools

	- name: Setup Bun
	# `:trailblaze-desktop:releaseArtifacts` transitively triggers
	# `:trailblaze-scripting-subprocess:installTrailblazeScriptingSdk`,
	# which runs `bun install` to lay down the scripting SDK's
	# node_modules ahead of JAR packaging.
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Build trailblaze uber JAR (with WASM template embedded)
	# `install-trailblaze-source.sh` invokes
	# `:trailblaze-desktop:releaseArtifacts`, which runs
	# `packageUberJarForCurrentOS` and copies the launcher script
	# alongside the JAR into `trailblaze-desktop/build/release/`.
	# TRAILBLAZE_GRADLE_EXTRA_ARGS layers in `-Ptrailblaze.wasm=true` so
	# the report template gets bundled into the JAR's classpath, and
	# `--parallel` so independent modules compile concurrently (the
	# `checks` job already runs the same task graph with `--parallel`).
	env:
	TRAILBLAZE_GRADLE_EXTRA_ARGS: -Ptrailblaze.wasm=true --parallel
	run: ./scripts/install-trailblaze-source.sh

	- name: Stage artifact contents
	# `releaseArtifacts` lays down the JAR + launcher with stable names
	# at `trailblaze-desktop/build/release/`. Copy them to a flat
	# top-level staging dir so the upload step keeps the layout simple
	# — `actions/download-artifact@v4` extracts into a directory whose
	# contents match what was uploaded, and the install-from-artifact
	# script expects `trailblaze.jar` + `trailblaze` side-by-side.
	run: \|
	mkdir -p trailblaze-uber-jar
	cp trailblaze-desktop/build/release/trailblaze.jar trailblaze-uber-jar/
	cp trailblaze-desktop/build/release/trailblaze trailblaze-uber-jar/
	ls -lh trailblaze-uber-jar/

	- name: Upload uber JAR artifact
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-uber-jar
	path: trailblaze-uber-jar/
	if-no-files-found: error
	retention-days: 7

	# Run Tests. Split into two parallel jobs so the slow
	# `:trailblaze-scripting-subprocess` test suite (~45min — spawns a `bun`/`tsx`
	# subprocess per test) doesn't gate the rest of the build behind a single
	# 60-minute timeout. Wall-clock time becomes max(jobs), not sum.
	checks:
	runs-on: ubuntu-latest
	timeout-minutes: 30

	steps:
	- name: Checkout
	uses: actions/checkout@v4

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: 'zulu'
	java-version: '17'
	check-latest: true

	- name: Setup Gradle
	# Dependency + build cache reuse across runs (seeded on main pushes,
	# restored read-only on PRs) — unchanged modules restore compiled
	# classes and test results metadata from cache instead of rebuilding.
	uses: gradle/actions/setup-gradle@9c971963bec38e04b3d30dcc455b5382be2fdbfb # v6.3.0
	with:
	cache-read-only: ${{ github.ref != 'refs/heads/main' }}

	- name: Setup Bun
	# bun is the sole supported JS runtime. `assemble check` triggers
	# `:trailblaze-scripting-subprocess:installTrailblazeScriptingSdk`
	# (a `bun install` that downstream JAR builds read) and the host
	# `CheckCommandTest` suite, which shells out to the bun-test phase.
	# Without bun on PATH both fail hard rather than skipping.
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Install ffmpeg
	# The trailblaze-capture video tests encode real recordings; see the script for
	# why it is a pinned build.
	run: ./.github/install-ffmpeg.sh

	- name: Test on Ubuntu (excluding scripting-subprocess)
	# `--parallel` lets independent modules compile and test concurrently
	# across the daemon's worker pool; combined with carving out the slow
	# subprocess suite into its own job this brings the wall-clock down
	# from 60+ minutes to roughly 10–15.
	run: >-
	./gradlew --parallel assemble check
	-x :trailblaze-scripting-subprocess:test
	-x :trailblaze-scripting-subprocess:check

	- name: Verify Documentation is up-to-date
	run: ./scripts/generate-docs-and-diff.sh

	# Isolated job for the subprocess-spawning tests. Each test forks a
	# `bun`/`tsx` child process to exercise the MCP transport, so the suite
	# is i/o-bound and serializes inside a single JVM — running it in its own
	# CI job lets the faster `checks` job report status promptly even when
	# this one is still working through its subprocess fan-out.
	checks-scripting-subprocess:
	runs-on: ubuntu-latest
	timeout-minutes: 60

	steps:
	- name: Checkout
	uses: actions/checkout@v4

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: 'zulu'
	java-version: '17'
	check-latest: true

	- name: Setup Gradle
	# Dependency + build cache reuse across runs (seeded on main pushes,
	# restored read-only on PRs).
	uses: gradle/actions/setup-gradle@9c971963bec38e04b3d30dcc455b5382be2fdbfb # v6.3.0
	with:
	cache-read-only: ${{ github.ref != 'refs/heads/main' }}

	- name: Setup Bun
	# The subprocess tests `bun install` and then spawn `bun`/`tsx` for each
	# MCP server fixture. The other Android jobs already set this up; the
	# checks job didn't need it before, but this isolated suite does.
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Run scripting-subprocess tests
	run: ./gradlew :trailblaze-scripting-subprocess:check

	# PR-time web showcase validation shares the Linux uber JAR built above.
	# The standalone Wikipedia workflow stays push-only so `main` still produces
	# the docs gallery artifact that GitHub Pages publishes.
	wikipedia-trails:
	if: github.event_name == 'pull_request'
	runs-on: ubuntu-latest
	timeout-minutes: 60
	needs: build-uber-jar
	env:
	OPENAI_API_KEY: no-key-set
	# Record each session's screen so the report plays the recording (off by default).
	TRAILBLAZE_CAPTURE_VIDEO: "true"

	steps:
	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Bun
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Cache Playwright driver
	# The uber JAR ships without Playwright's ~200 MB driver-bundle; the CLI
	# downloads it from Maven Central at first use and caches it under
	# ~/.cache/trailblaze/playwright-driver. Restore it across runs so the
	# trail isn't exposed to that download (a stalled Maven Central
	# connection here has burned a full 10-minute no-progress timeout).
	# The main-seeded entry comes from wikipedia-trails.yml, since caches
	# created inside a PR are only visible to that PR.
	uses: actions/cache@v4
	with:
	path: ~/.cache/trailblaze/playwright-driver
	key: playwright-driver-${{ runner.os }}-${{ hashFiles('gradle/libs.versions.toml') }}

	- name: Download prebuilt uber JAR
	uses: actions/download-artifact@v4
	with:
	name: trailblaze-uber-jar
	path: trailblaze-uber-jar/

	- name: Install trailblaze from prebuilt artifact
	env:
	TRAILBLAZE_ARTIFACT_DIR: ${{ github.workspace }}/trailblaze-uber-jar
	run: ./scripts/install-trailblaze-from-artifact.sh

	- name: Install Playwright system dependencies
	run: \|
	sudo apt-get update
	sudo apt-get install -y \
	ffmpeg \
	libasound2t64 \
	libatk-bridge2.0-0 \
	libatk1.0-0 \
	libcups2 \
	libdrm2 \
	libgbm1 \
	libgtk-3-0 \
	libnss3 \
	libxcomposite1 \
	libxdamage1 \
	libxfixes3 \
	libxkbcommon0 \
	libxrandr2

	- name: Install ffmpeg
	# Session video encodes through it; see the script for why it is a pinned build.
	run: ./.github/install-ffmpeg.sh

	- name: Create and run Wikipedia trail
	id: run-trails
	continue-on-error: true
	run: ./.github/pr_run_wikipedia_trails.sh

	- name: Generate Trailblaze report
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_trailblaze_report.sh

	- name: Generate docs report-gallery assets
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_report_assets.sh

	- name: Create artifacts
	if: always()
	continue-on-error: true
	run: ./.github/pr_create_artifacts.sh

	- name: Upload artifacts
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-artifacts-wikipedia
	path: \|
	trailblaze-logs.zip
	trailblaze_report.html

	- name: Upload docs report-gallery assets
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: docs-report-wikipedia
	path: report-assets/
	if-no-files-found: warn

	- name: Fail if trails failed
	if: steps.run-trails.outcome == 'failure'
	run: \|
	echo "Wikipedia trails failed - failing the workflow"
	exit 1

	# PR-time iOS showcase validation shares the same uber JAR built above.
	# The standalone iOS workflow stays push-only so `main` still produces the
	# docs gallery artifact that GitHub Pages publishes.
	#
	# Runs on macos-26 (Xcode 26 / iOS 26 simulators, iPhone 17 Pro) to match the
	# showcase recording in docs/showcase-trails.yml — the scripted Contacts tools
	# drive the iOS 26 Contacts UI. Kept in sync with .github/workflows/ios-contacts-trails.yml.
	ios-contacts-trails:
	if: github.event_name == 'pull_request'
	runs-on: macos-26
	timeout-minutes: 60
	needs: build-uber-jar
	env:
	OPENAI_API_KEY: no-key-set
	# Record each session's screen so the report plays the recording (off by default).
	TRAILBLAZE_CAPTURE_VIDEO: "true"

	steps:
	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Bun
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Download prebuilt uber JAR
	uses: actions/download-artifact@v4
	with:
	name: trailblaze-uber-jar
	path: trailblaze-uber-jar/

	- name: Install trailblaze from prebuilt artifact
	env:
	TRAILBLAZE_ARTIFACT_DIR: ${{ github.workspace }}/trailblaze-uber-jar
	run: ./scripts/install-trailblaze-from-artifact.sh

	- name: Boot iOS Simulator
	run: \|
	echo "Available iPhone simulators:"
	xcrun simctl list devices available \| grep "iPhone"

	# Prefer iPhone 17 Pro (iOS 26 runtime on macos-26 — the device the
	# showcase recording was captured on); fall back to the first iPhone.
	SIMULATOR_UDID=$(
	xcrun simctl list devices available \|
	grep "iPhone 17 Pro " \| grep -v "Max\\|Plus" \|
	head -1 \|
	grep -oE "[A-F0-9]{8}-[A-F0-9]{4}-[A-F0-9]{4}-[A-F0-9]{4}-[A-F0-9]{12}"
	)

	if [ -z "$SIMULATOR_UDID" ]; then
	echo "iPhone 17 Pro not found, falling back to first available iPhone..."
	SIMULATOR_UDID=$(
	xcrun simctl list devices available \|
	grep "iPhone" \|
	head -1 \|
	grep -oE "[A-F0-9]{8}-[A-F0-9]{4}-[A-F0-9]{4}-[A-F0-9]{4}-[A-F0-9]{12}"
	)
	fi

	if [ -z "$SIMULATOR_UDID" ]; then
	echo "ERROR: No iPhone simulator found"
	exit 1
	fi

	echo "Booting simulator UDID: $SIMULATOR_UDID"
	xcrun simctl boot "$SIMULATOR_UDID"
	xcrun simctl bootstatus "$SIMULATOR_UDID" -b

	echo "SIMULATOR_UDID=$SIMULATOR_UDID" >> "$GITHUB_ENV"
	echo "Simulator booted: $SIMULATOR_UDID"

	- name: Install ffmpeg (session video + animated-timeline WebP)
	continue-on-error: true
	run: brew install ffmpeg

	- name: Run iOS Contacts trail
	id: run-trails
	continue-on-error: true
	run: ./.github/pr_run_ios_contacts_trails.sh

	- name: Generate Trailblaze report
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_trailblaze_report.sh

	- name: Generate docs report-gallery assets
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_report_assets.sh

	- name: Create artifacts
	if: always()
	continue-on-error: true
	run: ./.github/pr_create_artifacts.sh

	- name: Upload artifacts
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-artifacts-ios-contacts
	path: \|
	trailblaze-logs.zip
	trailblaze_report.html

	- name: Upload docs report-gallery assets
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: docs-report-ios-contacts
	path: report-assets/
	if-no-files-found: warn

	- name: Fail if trails failed
	if: steps.run-trails.outcome == 'failure'
	run: \|
	echo "iOS Contacts trails failed - failing the workflow"
	exit 1

	# Drives the trail through `connectedDebugAndroidTest` — the test logic runs on the
	# emulator inside the JUnit instrumentation process and talks to the headless desktop
	# server on the default HTTPS port (TrailblazeDevicePort.TRAILBLAZE_DEFAULT_HTTPS_PORT) for agent orchestration.
	android-tests-on-device:
	runs-on: ubuntu-latest
	timeout-minutes: 60
	env:
	# These CI checks replay recorded trails, so LLM inference is never invoked.
	# A non-empty sentinel value is required to satisfy the provider/client wiring
	# (e.g., `examples/build.gradle.kts` forwards this to `trailblaze.llm.auth.token.openai`
	# and the on-device client map is built from whichever tokens are present).
	OPENAI_API_KEY: no-key-set

	steps:
	- name: Enable KVM group perms
	if: runner.os == 'Linux'
	run: \|
	echo 'KERNEL=="kvm", GROUP="kvm", MODE="0666", OPTIONS+="static_node=kvm"' \| sudo tee /etc/udev/rules.d/99-kvm4all.rules
	sudo udevadm control --reload-rules
	sudo udevadm trigger --name-match=kvm
	ls /dev/kvm

	- name: Keep the daemon's ports out of the ephemeral range
	# 52525/52526 fall inside Linux's ephemeral range (32768-60999), and the emulator and its
	# helpers bind random ports from it. One landing on 52525 first makes the daemon exit with
	# "Address already in use". Reserving them keeps the kernel from handing them out.
	if: runner.os == 'Linux'
	run: sudo sysctl -w net.ipv4.ip_local_reserved_ports=52525-52526

	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Gradle
	# Dependency + build cache reuse across runs (seeded on main pushes,
	# restored read-only on PRs). This job runs Gradle twice — the
	# instrumentation build below and the WASM-enabled report fallback in
	# pr_generate_trailblaze_report.sh — both benefit.
	uses: gradle/actions/setup-gradle@9c971963bec38e04b3d30dcc455b5382be2fdbfb # v6.3.0
	with:
	cache-read-only: ${{ github.ref != 'refs/heads/main' }}

	- name: Setup Bun
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Create and start emulator and run tests (Linux)
	if: runner.os == 'Linux'
	uses: reactivecircus/android-emulator-runner@b530d96654c385303d652368551fb075bc2f0b6b
	id: run-tests-linux
	continue-on-error: true
	with:
	api-level: 34
	arch: x86_64
	disable-animations: true
	disk-size: 6000M
	heap-size: 600M
	profile: Nexus 6
	script: ./.github/pr_run_android_tests_on_device.sh

	- name: Verify logs were pulled
	if: always()
	run: ./.github/pr_verify_logs_were_pulled.sh

	- name: Generate Trailblaze report
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_trailblaze_report.sh

	- name: Create artifacts
	if: always()
	continue-on-error: true
	run: ./.github/pr_create_artifacts.sh

	- name: Upload artifacts
	if: always() # Ensures this runs even if the previous step failed
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-artifacts-on-device
	path: \|
	trailblaze-logs.zip
	trailblaze_report.html
	logcat.log

	- name: Fail if tests failed
	if: steps.run-tests-linux.outcome == 'failure'
	run: \|
	echo "Tests failed - failing the workflow"
	exit 1

	# Drives the trail from the host via the Trailblaze CLI (`trailblaze trail …`).
	# Agent loop runs in-process on the host; tools dispatch over RPC to the bundled
	# on-device APK that the CLI auto-installs over ADB.
	android-tests-host-rpc:
	runs-on: ubuntu-latest
	timeout-minutes: 60
	needs: build-uber-jar
	env:
	# Recordings-only — LLM is never invoked. See android-tests-on-device above.
	OPENAI_API_KEY: no-key-set
	# Record each session's screen so the report plays the recording (off by default).
	TRAILBLAZE_CAPTURE_VIDEO: "true"

	steps:
	- name: Enable KVM group perms
	if: runner.os == 'Linux'
	run: \|
	echo 'KERNEL=="kvm", GROUP="kvm", MODE="0666", OPTIONS+="static_node=kvm"' \| sudo tee /etc/udev/rules.d/99-kvm4all.rules
	sudo udevadm control --reload-rules
	sudo udevadm trigger --name-match=kvm
	ls /dev/kvm

	- name: Keep the daemon's ports out of the ephemeral range
	# 52525/52526 fall inside Linux's ephemeral range (32768-60999), and the emulator and its
	# helpers bind random ports from it. One landing on 52525 first makes the daemon exit with
	# "Address already in use". Reserving them keeps the kernel from handing them out.
	if: runner.os == 'Linux'
	run: sudo sysctl -w net.ipv4.ip_local_reserved_ports=52525-52526

	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Android SDK
	uses: android-actions/setup-android@9fc6c4e9069bf8d3d10b2204b1fb8f6ef7065407
	with:
	# Drop the action's default `tools` package. Google removed the obsolete SDK Tools
	# package from the repository, so `sdkmanager tools` fails outright on runner images
	# that no longer ship it preinstalled. `platform-tools` is what the build needs.
	packages: platform-tools

	- name: Setup Bun
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Download prebuilt uber JAR
	uses: actions/download-artifact@v4
	with:
	name: trailblaze-uber-jar
	path: trailblaze-uber-jar/

	- name: Install trailblaze from prebuilt artifact
	env:
	TRAILBLAZE_ARTIFACT_DIR: ${{ github.workspace }}/trailblaze-uber-jar
	run: ./scripts/install-trailblaze-from-artifact.sh

	- name: Install ffmpeg
	# Session video encodes through it; see the script for why it is a pinned build.
	run: ./.github/install-ffmpeg.sh

	- name: Create and start emulator and run tests (Linux)
	if: runner.os == 'Linux'
	uses: reactivecircus/android-emulator-runner@b530d96654c385303d652368551fb075bc2f0b6b
	id: run-tests-linux
	continue-on-error: true
	with:
	api-level: 34
	arch: x86_64
	disable-animations: true
	disk-size: 6000M
	heap-size: 600M
	profile: Nexus 6
	script: ./.github/pr_run_android_tests_host_rpc.sh

	- name: Verify logs were pulled
	if: always()
	run: ./.github/pr_verify_logs_were_pulled.sh

	- name: Generate Trailblaze report
	if: always()
	continue-on-error: true
	run: ./.github/pr_generate_trailblaze_report.sh

	- name: Create artifacts
	if: always()
	continue-on-error: true
	run: ./.github/pr_create_artifacts.sh

	- name: Upload artifacts
	if: always() # Ensures this runs even if the previous step failed
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-artifacts-host-rpc
	path: \|
	trailblaze-logs.zip
	trailblaze_report.html
	logcat.log

	- name: Fail if tests failed
	if: steps.run-tests-linux.outcome == 'failure'
	run: \|
	echo "Tests failed - failing the workflow"
	exit 1

	# Drives a first-class TrailblazeTool over the daemon's `POST /mcp` endpoint —
	# the surface an external MCP client (Claude Code, Codex, Goose) uses. This is
	# the only job that reaches `TrailblazeMcpBridgeImpl.executeToolViaRpc` with
	# the interface's `blocking = false` default: android-tests-on-device is pure
	# instrumentation, and android-tests-host-rpc goes through DesktopYamlRunner.
	# Asserts an on-device failure surfaces as `isError` and a success returns the
	# tool's own output rather than an outcome-free placeholder.
	android-tests-mcp-dispatch:
	runs-on: ubuntu-latest
	timeout-minutes: 60
	needs: build-uber-jar
	env:
	# Tools are dispatched by name — the inner agent (and therefore the LLM) is
	# never invoked. See android-tests-on-device above.
	OPENAI_API_KEY: no-key-set
	# Record each session's screen so the report plays the recording (off by default).
	TRAILBLAZE_CAPTURE_VIDEO: "true"

	steps:
	- name: Enable KVM group perms
	if: runner.os == 'Linux'
	run: \|
	echo 'KERNEL=="kvm", GROUP="kvm", MODE="0666", OPTIONS+="static_node=kvm"' \| sudo tee /etc/udev/rules.d/99-kvm4all.rules
	sudo udevadm control --reload-rules
	sudo udevadm trigger --name-match=kvm
	ls /dev/kvm

	- name: Keep the daemon's ports out of the ephemeral range
	# 52525/52526 fall inside Linux's ephemeral range (32768-60999), and the emulator and its
	# helpers bind random ports from it. One landing on 52525 first makes the daemon exit with
	# "Address already in use". Reserving them keeps the kernel from handing them out.
	if: runner.os == 'Linux'
	run: sudo sysctl -w net.ipv4.ip_local_reserved_ports=52525-52526

	- name: Checkout
	uses: actions/checkout@v4
	with:
	fetch-depth: 0

	- name: Setup Java
	uses: actions/setup-java@v4
	with:
	distribution: zulu
	java-version: 17

	- name: Setup Android SDK
	uses: android-actions/setup-android@9fc6c4e9069bf8d3d10b2204b1fb8f6ef7065407
	with:
	# Drop the action's default `tools` package. Google removed the obsolete SDK Tools
	# package from the repository, so `sdkmanager tools` fails outright on runner images
	# that no longer ship it preinstalled. `platform-tools` is what the build needs.
	packages: platform-tools

	- name: Setup Bun
	uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0

	- name: Download prebuilt uber JAR
	uses: actions/download-artifact@v4
	with:
	name: trailblaze-uber-jar
	path: trailblaze-uber-jar/

	- name: Install trailblaze from prebuilt artifact
	env:
	TRAILBLAZE_ARTIFACT_DIR: ${{ github.workspace }}/trailblaze-uber-jar
	run: ./scripts/install-trailblaze-from-artifact.sh

	- name: Install ffmpeg
	# Session video encodes through it; see the script for why it is a pinned build.
	run: ./.github/install-ffmpeg.sh

	- name: Create and start emulator and run MCP dispatch check (Linux)
	if: runner.os == 'Linux'
	uses: reactivecircus/android-emulator-runner@b530d96654c385303d652368551fb075bc2f0b6b
	id: run-tests-linux
	continue-on-error: true
	with:
	api-level: 34
	arch: x86_64
	disable-animations: true
	disk-size: 6000M
	heap-size: 600M
	profile: Nexus 6
	script: ./.github/pr_run_android_mcp_dispatch.sh

	- name: Create artifacts
	if: always()
	continue-on-error: true
	run: ./.github/pr_create_artifacts.sh

	- name: Upload artifacts
	if: always() # Ensures this runs even if the previous step failed
	uses: actions/upload-artifact@v4
	with:
	name: trailblaze-artifacts-mcp-dispatch
	path: \|
	trailblaze-logs.zip
	logcat.log
	mcp-dispatch/

	- name: Fail if the MCP dispatch check failed
	if: steps.run-tests-linux.outcome == 'failure'
	run: \|
	echo "MCP dispatch check failed - failing the workflow"
	exit 1

Navigation Menu

Record video on every CI trail run, and keep recordings in sync on older ffmpeg #773

Workflow file

Record video on every CI trail run, and keep recordings in sync on older ffmpeg #773

Uh oh!

Workflow file for this run