Content
96%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 exceptionally tight, operator-grade runbook: executable commands with exact paths, an error-to-action preflight table, explicit polling and verification checkpoints, and gotchas that are genuinely non-obvious environment knowledge. The only structural nit is that the single SKILL.md inlines three procedure variants and a gotchas list that could be split into one-level-deep reference files.
Suggestions
Move the 'On a Windows host (no UTM)' alternate transport flow into a references/windows-host.md (or similar) one-level-deep file, keeping SKILL.md as the UTM-first overview.
Consider splitting the Gotchas list and the preflight error table into a short reference file so the main body stays a lean overview of the core workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and every token carries non-obvious environment knowledge Claude would not have: the preflight output table, "zsh `echo` eats backslashes in Windows paths", "A PowerShell 5 script without a BOM is read as the system code page", and "the Homebrew symlink reports 'Application not found'". There is no padding or explanation of known concepts, so it sits at the top anchor rather than 4. | 5 / 5 |
Actionability | Guidance is fully executable with copy-paste-ready commands and exact paths, e.g. `scripts/windows/utm.sh task sstest-full 'C:\Users\Public\sstest\e2e-file-links.ps1' -- -Exe C:\Users\Public\sstest\ss.exe ... -Extended` and the one-liner `powershell -ExecutionPolicy Bypass -File C:\Users\Public\ss-setup.ps1`. Specific examples cover the common cases (auto, manual, native Windows host). | 5 / 5 |
Workflow Clarity | The sequence Preflight → Build → 3a/3b → clean → report has explicit validation and error-recovery feedback loops: the probe table maps each failure output to an action (`OSStatus -1712` → ask user to allow, `desktopUser=` empty → ask user to log in), step 3 polls the output file until the last line is `DONE`, step 4 verifies with `fsutil reparsepoint query` reparse tags, and results must be reported per token with the exact failing output. Not a 4: checkpoints and recovery loops are present end-to-end, including cleanup via `utm.sh clean`. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/ in the skill), so this is a single-file skill with well-organized sections (Rules, Preflight, Build, 3a, 3b, On a Windows host, Gotchas) and easy navigation. It falls between anchors 4 and 5: at ~107 lines with three distinct procedure variants (UTM auto, UTM manual, native Windows host), some content — notably the 'On a Windows host (no UTM)' alternate flow and the Gotchas list — could live in a one-level-deep reference file, which is the 'minor organization gap' of anchor 4 rather than the fully split structure of anchor 5. | 4 / 5 |
Total | 19 / 20 Passed |