Content
63%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 delivers strong, concrete Windmill/DuckDB guidance with real validation checkpoints around preview, metadata regeneration, and deploy gating — clearly actionable. Its weaknesses are repetition (deploy caution stated three times, over-dense bullets) and the inlining of the general Windmill script-writing guide, which should be split into a reference file so SKILL.md stays a focused overview for the DuckDB dialect. No bundle files exist, so all content is carried in the main file.
Suggestions
State the deploy-only-on-explicit-ask rule once (in the CLI command list) and reference it from the other sections instead of restating it three times.
Move the 'Windmill Script Writing Guide' (General Principles, Return Values, Preprocessor Scripts) into a references/ file (e.g. references/windmill-script-writing.md) and keep a short overview with a clearly signaled link in SKILL.md, keeping the main file focused on the DuckDB dialect.
Show one complete concrete preview invocation (e.g. wmill script preview f/my/script -d '{"name": "alice", "age": 30}') so the argument-passing shape is copy-paste ready rather than implied.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The deploy-only-on-explicit-ask caution is repeated three times (the CLI list, 'Preview vs run', and the 'After writing' section), and several bullets are dense multi-clause paragraphs (the custom-resource-type bullet, the metadata-sync paragraph) that could be tightened. Most content is non-obvious Windmill-specific knowledge rather than filler, so this lands at 'Mostly efficient but includes some unnecessary explanation or could be tightened', not the padded verbosity of score 2. | 3 / 5 |
Actionability | Concrete, runnable commands appear throughout ('wmill script preview <script_path>', 'wmill generate-metadata --dry-run', 'wmill generate-metadata rehash', 'wmill resource-type list --schema') and the SQL blocks are executable ('ATTACH \'ducklake\' AS dl;', 'COPY (SELECT * FROM users) TO \'s3:///exports/users.parquet\' (FORMAT PARQUET)'). Minor gaps keep it below fully copy-paste-ready: the '-d \'<args>\'' preview invocation is described but never shown as a concrete example, and placeholder syntaxes for other dialects are listed where DuckDB is the skill's focus. | 4 / 5 |
Workflow Clarity | The write → preview → generate-metadata → deploy-only-when-asked flow has real decision rules (preview vs run chosen by intent, sync push only on explicit ask) and genuine validation checkpoints: preview validates before any deploy, 'diff the regenerated .lock / .script.lock files and tell the user which dependency versions changed', and '--dry-run ... lists each stale item with a reason'. It falls short of score 5 because the sequence is scattered across four separate sections rather than presented as one coherent ordered workflow with error-recovery loops. | 4 / 5 |
Progressive Disclosure | The skill is ~150 lines with no bundle files, and the entire general 'Windmill Script Writing Guide' (General Principles, Return Values, Preprocessor Scripts) is inlined — content that belongs in a separate reference file for a skill named write-script-duckdb. Navigation points outward to project files ('see the Deploying section in AGENTS.wmill.md', the 'preview' skill) rather than to organized bundle references, matching 'content that should be separate is inline' rather than the well-split structure of score 4. | 3 / 5 |
Total | 14 / 20 Passed |