Content
77%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 content is an exceptionally actionable, well-validated guide to building Windmill raw apps, with copy-paste commands, executable code, guarded destructive operations, and rare insider gotchas (non-TTY wizard hangs, draft-vs-deployed runnables, unmounted-entrypoint blank screens). Its weaknesses are structural: two full guides are concatenated into one ~590-line file with duplicated setup, metadata, and roles content, and no reference files despite explicitly describing a companion guide. Splitting the authoring material into a reference file and deduplicating would raise both conciseness and progressive disclosure.
Suggestions
Split the inlined companion guide (everything from the second "# Windmill Raw Apps" heading onward) into a references/ file (e.g. references/authoring.md) and keep SKILL.md as the CLI-workflow overview that links to it one level deep, as line 8 already implies the structure should be.
Deduplicate the repeated guidance: the "Data setup" questioning rules, the `wmill generate-metadata` workflow, and the `data.roles` semantics each appear in both halves — state each once and cross-reference it.
Isolate version-sensitive specifics (React 19 / React 18 / Svelte 5 defaults, Tailwind CSS v4 directives) in one place so future version churn doesn't scatter edits across the whole document.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows and is dense with domain-specific value, but it duplicates material across its two concatenated halves: "Data setup" guidance appears both in "Settle the setup" and again in the second document's "Data setup" section, the `wmill generate-metadata` workflow is explained twice, and the `data.roles` semantics appear in both the CLI and authoring halves. Anchor 3 ("mostly efficient but could be tightened") fits; not 4 because the duplication is wholesale rather than minor. | 3 / 5 |
Actionability | Fully executable throughout: exact commands with flags (`wmill app new --summary ... --framework react19 --datatable main --schema app1`), runnable TypeScript/Python/YAML/SQL examples, a precise bindings table (`backend.<key>(args)`, `backendAsync`, `waitJob`, `getJob`, `streamJob`), and copy-paste entrypoint snippets. This matches anchor 5 ("copy-paste ready code or commands; specific examples cover the common cases"). | 5 / 5 |
Workflow Clarity | App creation is explicitly sequenced (Step 1 settle setup → Step 2 run the command → Step 3 offer preview) with an anti-patterns list, the migration workflow ends with the required follow-up ("add them to data.tables"), and validation checkpoints are explicit: run `wmill app lint` after edits, check the regenerated `.lock` diff and report dependency changes, and confirm before `--overwrite`. Destructive operations are guarded, so the missing-validation cap does not apply; this matches anchor 5. | 5 / 5 |
Progressive Disclosure | The body announces "the companion authoring guide" (line 8) — which turns out to be inlined in full as a second "# Windmill Raw Apps" H1 starting at line 250, so two complete documents share one SKILL.md with no references/ bundle at all. Section headers are good, but ~340 lines that clearly belong in a separate reference file are inlined; anchor 3 ("content that should be separate is inline") fits, and it is not 4 because the split is over half the file, not a minor gap. | 3 / 5 |
Total | 16 / 20 Passed |