[Brian Krabach (bkrabach)]

Include the portable UI-use behavior once in the main Work bundle so work-amp-dev inherits it automatically. Ordinary model selection remains the default, and optional decisions reuse explicitly enabled provider-owned backends plus per-surface consent. Reusable composition preserves the host's orchestrator, context and provider choices.

The default behavior installs no browser SDK, decision SDK or desktop executor. A host supplies its connected/managed browser; direct browser execution needs the optional browser package and host setup. Desktop execution remains an explicit behaviors/computer.yaml/canonical computer-use opt-in. The work-ui-use alias adds no second tool or replacement runtime.

Merge prerequisites: UI-use main, streaming #68, and live #19. Anonymous canonical UI-use main resolution now passes. Optional desktop use additionally benefits from computer-use #31, which is not required for connected-browser use.

Validation in an isolated DTU with an explicit candidate UI-use source override: tests/test_decision_composition.py, tests/test_composition.py, and tests/test_amp_dev_composition.py — 22 passed. Checks verify one UI-use tool, ordinary defaults, explicit desktop opt-in, and preservation of host runtime/provider configuration. No source URL is pinned to a test revision, and this fixture composition evidence is not a fresh public-install or live-site acceptance claim.

Public-source qualification: the same 22 focused composition tests passed without candidate overrides, and the normal GitHub composition workflow passed all 98 tests. Streaming/live capability prerequisites above remain required for bounded execution.

Review candidate: 0771b3d522754944ca1b84cc9a7f135df458008e. Normal CI: https://github.com/microsoft/amplifier-bundle-work/actions/runs/39719191835 . Public UI-use main qualified at a1d3016faf238122f4fbeff809da7f6fe98bde9a. Ready for review; do not merge before streaming #68 and live #19.

[Brian Krabach (bkrabach)]

The full Work browser path now passes in a fresh, isolated host and worker using anonymous public Git dependency resolution, with no development source overrides. This is pre-merge candidate qualification, using an explicitly frozen Unified candidate wheel; it is not evidence that these changes have landed on main.

Two normal chats completed the synthetic form, with independent page-state verification:

• Ordinary default: three ordinary selections, no decision binding or mounted decision provider, despite the optional package and credential being available.
• Explicit Jev configuration: an initial five-second decision timeout recovered through ordinary selection; Jev 1.14.0 then selected Team Blue and Save. No approval prompts or extra writes were needed.

Exact shared sources:

Repository	Qualified commit
amplifier-bundle-work	0771b3d522754944ca1b84cc9a7f135df458008e
amplifier-module-loop-streaming	20792f1cc7afe0945a789db61a3fa476306efd69
amplifier-module-loop-live	646707c7814c7a3cfef99c11195df89b991c108c
amplifier-module-provider-openai	2fe28bdcb65ab38418eaa7353094262ff9a13f9b
amplifier-bundle-ui-use	a1d3016faf238122f4fbeff809da7f6fe98bde9a
amplifier-module-provider-jev	7ccade1eed84b17f215bc2e80253a1fb4c3a2130
amplifier-lib-decisions	6c9e064acb93d4a6ed090ee58a111b8e77648784

Host imports, all 120 worker packages, and Work/UI-use bundle origins were recorded. Work-amp-dev's inheritance was verified from the exact source; it was not a third live chat. The application artifact was the frozen development candidate 09628861740e373d17a66c66524060e05e0b26b2, wheel SHA-256 0bede22b65ffcb54194f98ac9a9866eaab11bfc458dd9d55f55c4fb6f95ac065. Startup readiness and trace-validation harness corrections are retained in the evidence; neither live task was repeated to hide a product failure. Temporary environments were removed afterward.

Independent approval and the existing merge order remain required: Streaming, then Live's lock refresh/CI and merge, then Work. The final check will use the actual accepted main branches. These runs are qualification observations, not a controlled performance comparison; desktop, enterprise SSO, and native Mac execution are outside their scope.