Trigger Scenarios

需要队列PowerToys UI测试流水线诊断401/403权限或登录问题监控流水线运行状态及产物处理通过本地验证后的CI失败

Install

npx skills add microsoft/PowerToys --skill ui-tests-pipeline-ci -g -y

More Options

Non-standard path

npx skills add https://github.com/microsoft/PowerToys/tree/main/.github/skills/ui-tests-pipeline-ci -g -y

Use without installing

npx skills use microsoft/PowerToys@ui-tests-pipeline-ci

指定 Agent (Claude Code)

npx skills add microsoft/PowerToys --skill ui-tests-pipeline-ci -a claude-code -g -y

安装 repo 全部 skill

npx skills add microsoft/PowerToys --all -g -y

预览 repo 内 skill

npx skills add microsoft/PowerToys --list

SKILL.md

Frontmatter

{
    "name": "ui-tests-pipeline-ci",
    "license": "MIT",
    "description": "Microsoft FTE-only workflow for validating setup, queueing, monitoring, and stabilizing PowerToys UI Test Automation through an existing Azure CLI session and Azure DevOps REST APIs. Use after local VM suites pass, when asked to run UITests CI, perform a setup preflight\/readiness check, diagnose repeated az login prompts or 401\/403 permission failures, reuse a successful build, inspect recordings\/artifacts, or manage the three-run limit. Keywords: FTE, az, Azure CLI, Azure DevOps, pipeline, UI Test Automation, UITests CI, buildNow, specificBuildId, uiTestModules, CI flake."
}

PowerToys UI Tests Pipeline CI

Queue and stabilize the internal UI Test Automation Azure DevOps pipeline only after the target UITest suite is proven locally. Use the existing Azure CLI sign-in plus Azure DevOps REST APIs for discovery, preview, queueing, status, timelines, logs, tests, artifacts, and result attachments.

[!IMPORTANT] Microsoft FTE only. This workflow requires authorized access to the microsoft Azure DevOps organization and the Dart project. External contributors stop after local validation and report CI as unavailable. Never store credentials, PATs, tokens, or internal artifact contents in the repository.

When to use

Use this skill when an authorized Microsoft FTE asks to:

Queue PowerToys UITests in the internal UI Test Automation pipeline.
Validate Azure CLI and Azure DevOps readiness before queueing or after a 401/403 response.
Monitor a UITests pipeline run or summarize its stages, tests, and artifacts.
Iterate on a failure that passed the complete local VM matrix.
Reuse a prior successful product build while rebuilding only one or more UITest projects.
Find and share failed-test screenshots, logs, or recording links.

Do not use this skill for local execution. Complete ui-tests-local-vm first. Use ui-tests-migration for test implementation and stabilization.

Non-negotiable gates

Setup preflight first. Before the first Azure operation in a session, run Test-AzureDevOpsSetup.ps1 and require Ready=true with every required check PASS. It performs reads and a non-mutating pipeline preview only. Re-run it after account changes or any 401/403 response.
Local first. Do not queue CI until all required local runs are green, including full suites on the default and Constrained profiles for Windows 10 and Windows 11, plus the applicable architecture builds/guests required by ui-tests-local-vm.
Pushed revision. Queue only a pushed branch. Record its exact commit and verify the queued run's sourceVersion matches it.
One run per branch. Before queueing, discover active runs for UI Test Automation. Wait for or cancel a relevant superseded run on the target branch; runs on other branches may continue in parallel. Never cancel another branch's unrelated run.
Always scope modules. uiTestModules must be non-empty and contain the exact current UITest project stem, for example [FancyZonesEditor.UITests.Next].
Three-run ceiling. A CI stabilization sequence may queue at most three runs total. Keep an attempt ledger. If run 3 is not green, stop and ask the user for assistance. Also stop when three consecutive runs show no stabilization progress.
Evidence before edits. Read the failed result, logs, screenshot, and recording before forming a fix hypothesis. Preserve assertions and classify infrastructure failures separately.
Tracked runs remain unfinished work. After queueing, persist the build ID, branch, source SHA, attempt number, and parameters in session/task state. Do not mark the task complete or claim a terminal result while that build is nonterminal. If no authenticated completion waiter exists, arm the one-hour scheduled continuation in the agentic loop rather than relying on a passive handoff. On every scheduled wake, resume, notification, or user turn that continues the tracked CI task, query that exact build ID before other Azure work and continue from its current state.

Internal constants

Setting	Value
Azure DevOps organization	`microsoft`
Project	`Dart`
Pipeline name	`UI Test Automation`
Current known definition ID	`161438` (discover by name each session; do not blindly hardcode)
Azure DevOps token resource	`499b84ac-1321-427f-aa17-267ca6975798`
Required setup check	`scripts/Test-AzureDevOpsSetup.ps1`
Platforms	`arm64`, `x64`
Default booleans	`enableMsBuildCaching=false`, `useVSPreview=false`, `useLatestWebView2=false`

Required workflow

Read and execute references/agentic-loop.md from top to bottom. It contains:

The required prompt-free setup preflight and bundled REST helper.
Local-signoff and active-run preflight.
buildNow versus specificBuildId decision rules.
Exact queue parameters and branch targeting.
Monitoring, failure evidence, direct Azure Test attachment downloads, and recording links.
Tracked-run continuation, one-hour scheduled polling, and completion-notification limits.
The three-run stabilization ledger and stop conditions.

Completion standard

A task is complete only when one of these is true:

The run is terminal Succeeded, all selected tests executed, and there are no failed, aborted, timed-out, error, or not-executed results.
For a monitor-only request, the run is terminal but failed, and the report includes the controlling failure, relevant logs, and available recording/artifact links.
For a stabilization request, a terminal failed run is intermediate work. Continue the agentic loop until a later attempt succeeds, a genuine blocker prevents the next verified attempt, or the three-run/no-progress ceiling is reached.
The three-run ceiling or no-progress rule was reached, and the agent stopped and asked the user for assistance with the full attempt ledger and evidence links.