Merge branch 'main' into deps/lemonade-10.7.0 #101

	# NOTE: This workflow activates once #1297 (Strix Halo self-hosted runner) lands.
	# Until then it is validated out-of-band on Strix Halo class hardware.
	# The strix-halo runner is Windows. Python/uv setup goes through the
	# ./.github/actions/setup-venv composite action (the lab-runner convention,
	# see build_cpp.yml) which installs uv, creates .venv and puts it on PATH.
	# Inline `run:` steps use PowerShell — bash syntax (&&, \ continuation) is not
	# available there.
	name: Agent Behavior E2E

	on:
	workflow_dispatch:
	push:
	paths:
	- 'src/gaia/agents/builder/**'
	- 'src/gaia/agents/base/agent.py'
	- 'src/gaia/eval/behavior_harness.py'

	permissions:
	contents: read

	jobs:
	behavior-e2e:
	runs-on: [self-hosted, strix-halo]
	steps:
	- uses: actions/checkout@v6

	- name: Setup Python environment
	uses: ./.github/actions/setup-venv
	with:
	install-package: ".[dev]"

	- name: Ensure Lemonade
	shell: powershell
	run: \|
	gaia init --skip-lemonade
	if ($LASTEXITCODE -ne 0) { throw "gaia init failed" }

	- name: Run behavior E2E harness
	shell: powershell
	env:
	LEMONADE_BASE_URL: http://localhost:13305/api/v1
	run: \|
	python -m pytest tests/integration/eval/test_behavior_e2e.py `
	-m real_model -v --tb=short `
	--basetemp="${{ runner.temp }}\pytest-behavior"
	if ($LASTEXITCODE -ne 0) { throw "behavior E2E harness failed" }

	- name: Upload artifacts on failure
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: behavior-e2e-artifacts
	path: ${{ runner.temp }}/pytest-behavior//artifacts/

Provide feedback