Content
92%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
An exemplary operational skill: executable commands with full parameter sets, a validated multi-stage workflow with checkpoints and feedback loops, and clean one-level-deep reference navigation. The only real cost is token redundancy — the core gate rules are repeated across multiple sections and could be consolidated without losing compliance emphasis.
Suggestions
Consolidate the Windows-10-then-Windows-11 full-suite gate, currently stated in the intro, Guest OS policy, the mermaid cycle, task-list steps 10-11, and Non-negotiable rules: state it fully once and cross-reference from the other sections.
Merge the repeated 'never autopilot host setup / IsReady=false means BLOCKED' guidance (task-list step 0, Quick start, and Non-negotiable rules) into one authoritative section, keeping one-line pointers elsewhere.
Trim restatements of 'local green is not end-to-end completion' to the intro boundary statement plus the task-list step where the CI handoff actually happens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely project-specific (NTFS vs Dev Drive vmms wedging, DPAPI credential handling, LTSC build requirements) rather than explaining concepts Claude already knows, and all code blocks are commands, not exposition. But the Windows-10-then-Windows-11 full-suite gate and the 'local green is not completion / never autopilot host setup' rules are each restated across four to five sections (intro, Guest OS policy, the cycle, the task list, Non-negotiable rules), which is trim-able redundancy — more than a single minor instance, so not 5; the rest is efficient, so not 3. | 4 / 5 |
Actionability | Quotes: "pwsh .github\skills\ui-tests-local-vm\scripts\Initialize-LocalVmHost.ps1 -VmRoot X:\PowerToysUiTestVm -CheckOnly" and the full "Invoke-LocalVmUiTest.ps1" invocation with every parameter ("-Filter 'Name=MyModule.FocusedTest' -Platform x64Win10 -BuildLabel (git rev-parse HEAD) -SuiteTimeout 15m"), plus concrete practices like "read the controller result's `.Failed` array (non-passed tests + first error line) instead of re-parsing TRX". Copy-paste-ready commands cover the common cases, matching the top anchor. | 5 / 5 |
Workflow Clarity | The mermaid flowchart sequences design→build→package→run→evidence→widen→both-OS gate→constrained→CI with explicit decision nodes, and the 16-step checklist has real validation checkpoints: "Verify host setup FIRST ... If it reports IsReady=false, STOP", "Build product and test projects on the host to exit code 0", "Run the controller with -PlanOnly and inspect its request/plan", "Probe the non-admin interactive desktop before test execution", and "Always parse TRX and require `total > 0` plus `executed == total`". Feedback loops (focused test → diagnose first controlling failure → rebuild with -ReuseStagedPayload) are explicit, matching the top anchor exactly. | 5 / 5 |
Progressive Disclosure | "Required reads" maps each of the five real references/ files to a one-line scope with read conditions ("**read for any shell-extension module**", "Read only what the task needs"), all verified present; links are one level deep with correct section anchors (references/setup.md#0-... and #3-get-windows-media match actual setup.md headings), and scaffolded scripts (New-UiTestVm.ps1, Reset-LocalVm.ps1) are real targets documented in setup.md. The SKILL.md keeps only overview, policy, and the cycle while details live in the references — the top anchor's structure. | 5 / 5 |
Total | 19 / 20 Passed |