Content
71%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.
The body is highly actionable with exact wmill commands and copy-paste MSSQL examples, and the preview/metadata/deploy workflow carries genuine validation checkpoints. Its weaknesses are repeated deploy-vs-preview guidance, off-scope multi-language material inlined in an MSSQL-specific skill, and no use of bundle reference files for the generic Windmill content.
Suggestions
Consolidate the deploy/preview guidance stated in three places ("Deploy local changes", "Preview vs run", "After writing") into one section to remove redundancy.
Move the language-agnostic material (General Principles, Return Values, Preprocessor Scripts, non-MSSQL placeholder syntax) into a shared reference file and keep only the MSSQL section and CLI essentials in SKILL.md.
Reorder the sections chronologically (write → offer/run preview → generate-metadata → deploy) so the workflow reads as one sequence.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense platform-specific knowledge Claude would not know (wmill CLI semantics, lock/hash mechanics, @P1 conventions), not padded with known concepts, but it repeats the deploy/preview distinction in at least three sections ("Deploy local changes...", "Preview vs run", "After writing"), and in an MSSQL skill it inlines placeholder syntax for PostgreSQL/MySQL/Snowflake/BigQuery/Bash/PowerShell plus generic preprocessor material. That is more than the "minor instances of over-explanation" of anchor 4, matching anchor 3 ("Mostly efficient but includes some unnecessary explanation or could be tightened"). | 3 / 5 |
Actionability | Guidance is fully executable: exact commands (`wmill script preview <script_path>`, `wmill generate-metadata --dry-run`, `wmill generate-metadata rehash`, `wmill resource-type list --schema`) and copy-paste-ready SQL covering the common cases — @P1 parameter naming comments, `(s3object)` with OPENJSON WITH, and the `-- s3 prefix=... format=parquet` streaming directive. Anchor 5 ("Fully executable; copy-paste ready code or commands; specific examples cover the common cases") is a clear match. | 5 / 5 |
Workflow Clarity | The write → preview (validate) → generate-metadata → deploy-only-on-request workflow is clearly sequenced with real checkpoints: preview validates before any deploy, `--dry-run` lists stale items with reasons, and the lock diff/version-bump report gives a feedback loop. It falls short of anchor 5 because the flow is scattered across four non-chronological sections ("After writing — offer to test" appears after "Keep metadata in sync" though it precedes it in time) and there is no single coherent ordered procedure. Not anchor 3: validation for the risky operation (deploy overwriting the workspace version) is explicitly present, so the destructive-operation cap does not apply. | 4 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the ~119-line single file exceeds the under-50-lines simple-skill exception. The generic Windmill material (CLI commands, General Principles, Return Values, Preprocessor Scripts — covering TypeScript/Python/Go/Rust/etc.) is inline content that belongs in a shared reference file, while repeated pointers to an external `AGENTS.wmill.md` and "the `preview` skill" are not part of this bundle. This matches anchor 3 ("content that should be separate is inline") rather than anchor 2, since section headers and structure are genuinely good. | 3 / 5 |
Total | 15 / 20 Passed |