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 exceptionally actionable, with executable commands, code, and clearly sequenced workflows that include validation checkpoints and consent gates for risky operations. Its weaknesses are duplicated rule explanations across sections and a monolithic single-file structure that inlines a second guide the text itself describes as a companion. Splitting the authoring guide into a referenced file and de-duplicating the roles and data-setup rules would lift it further.
Suggestions
Split the inlined authoring guide (the second '# Windmill Raw Apps' section) into a separate reference file and point to it explicitly, instead of embedding two guides in one SKILL.md.
De-duplicate the rules stated twice: datatable roles (raw_app.yaml section vs. Data Tables rules 5-6) and the ask-first data setup ('Settle the setup' vs. 'Data setup') — state each once and cross-reference.
Trim restated framing such as the repeated deploy-vs-draft guidance across 'Draft code vs deployed code', the CLI Commands section, and Best Practices item 9.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and mostly high-value (flag tables, extension tables, anti-patterns, executable code), but the same rules are restated across sections — roles appear in both the raw_app.yaml section and Data Tables critical rules 5-6, and the ask-first data setup is explained in both 'Settle the setup' and 'Data setup' — going beyond the minor trimming of anchor 4. | 3 / 5 |
Actionability | Fully executable throughout: an exact `wmill app new` command with a flag table, complete entrypoint/backend/query code in TypeScript and Python, a bindings table with call signatures, and copy-paste YAML/SQL examples covering the common cases — matching the top anchor. | 5 / 5 |
Workflow Clarity | App creation is a clearly sequenced 3-step workflow with explicit validation checkpoints (single-round-trip questioning, 'Only proceed once every value is concrete', `wmill app lint` after edits, consent required for `--overwrite`, lock-diff review after `generate-metadata`, browser modal confirming migrations), with feedback loops like re-asking on ambiguous replies. | 5 / 5 |
Progressive Disclosure | There are no bundle files or references at all, and the text treats the authoring guide as a separate companion ('covered in the companion authoring guide') while inlining it in the same ~590-line file under a second H1 — good section structure, but content that clearly belongs in a separate file is inline, matching anchor 3. | 3 / 5 |
Total | 16 / 20 Passed |