Content
86%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.
A well-engineered skill body: executable examples across DABs/SDK/CLI, real and correctly anchored reference files, explicit compute routing, and a troubleshooting table. The remaining gains are tightening duplicated CLI listings and adding an explicit validate-fix-retry loop plus a caution on auto-approved destroys.
Suggestions
Collapse the "Common Operations" CLI listing and the "Development Workflow" steps into one section to remove the repeated validate/deploy/run commands.
Add an explicit feedback loop: "If bundle validate fails, fix the reported issue and re-validate before deploying."
Add a warning or confirmation step before `databricks bundle destroy --auto-approve`, since it irreversibly removes deployed resources.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost every section carries non-obvious specifics (run_if conditions, permission levels, the run-now --json idiom, "no cluster block = serverless"). Minor trimmable content exists — the full project-structure tree and the re-listing of CLI commands in both "Common Operations" and "Development Workflow" — but no padding explaining concepts Claude already knows, placing it above the midpoint, not at the lean-anchor 5. | 4 / 5 |
Actionability | Copy-paste-ready, complete examples in all three management modes: a non-interactive `databricks bundle init` heredoc, DABs YAML, a runnable Python SDK snippet with imports, and exact CLI commands including the non-obvious `jobs run-now --json '{"job_id": ...}'` form. The common cases (create, list, run, cancel, delete, validate, deploy) are all covered with real commands. | 5 / 5 |
Workflow Clarity | The Development Workflow gives a clear ordered sequence (validate → deploy → run → check status) with `bundle validate` as a pre-deploy checkpoint and a Common Issues table for error recovery. It stops short of the 5 anchor because there is no explicit validate→fix→retry loop, and the destructive `databricks bundle destroy --auto-approve` appears without any confirmation guidance. | 4 / 5 |
Progressive Disclosure | Textbook progressive disclosure: a Reference Files table maps use cases to four real one-level-deep files (all verified present), the task-type and trigger tables link to verified section anchors within those files, the body keeps only summaries inline, and routing to sibling skills (databricks-core, databricks-serverless) is clearly signaled at the top. | 5 / 5 |
Total | 18 / 20 Passed |