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: highly actionable, copy-paste-ready code across DABs, SDK, and CLI, with a clean reference-file table and anchor-level deep links to an existing four-file bundle. The workflow includes upfront validation, though it lacks an explicit fix-and-revalidate loop and a safety check before the destructive destroy command, and a few time-sensitive version strings (one contradicting the frontmatter) could be trimmed or consolidated.
Suggestions
Add a feedback loop to the Development Workflow (e.g., 'If validate fails, fix the config and re-run `databricks bundle validate` before deploying') and a verification step before `databricks bundle destroy --auto-approve`, which is destructive with no preceding check.
Trim the inline CLAUDE.md template's OS-specific CLI install instructions and reconcile the version references (body says '>= v0.288.0' while frontmatter compatibility says '>= v1.0.0'); move time-sensitive version pins like spark_version '15.4.x' to a dedicated or deprecated-patterns section to keep the body durable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and code/table-driven with almost no padding of concepts Claude already knows; nearly every section delivers executable configuration. Minor trimmable spots remain: the opening conceptual sentence ("Jobs orchestrate data workflows with multi-task DAGs, flexible triggers, and comprehensive monitoring"), an inline CLAUDE.md template with OS-specific CLI install instructions, and hardcoded time-sensitive versions (spark_version "15.4.x", CLI ">= v0.288.0" which also contradicts the frontmatter's ">= v1.0.0") that are not in a deprecated/old-patterns section. This is the score-4 'efficient; minor instances of over-explanation that could be trimmed' fit rather than the lean score-5 anchor. | 4 / 5 |
Actionability | Guidance is fully executable throughout: a non-interactive `databricks bundle init` command with a ready config-file heredoc, copy-paste YAML for DABs job definitions, complete Python SDK snippets (create/list/get/run_now/cancel/delete), exact CLI command lines including the non-obvious `run-now --json '{"job_id": ...}'` form, and a four-step deploy workflow. These cover the common cases in all three management surfaces, matching the score-5 'fully executable; copy-paste ready; specific examples cover the common cases' anchor. | 5 / 5 |
Workflow Clarity | The Development Workflow section sequences validate → deploy → run → check status with an explicit validation checkpoint (`databricks bundle validate`) up front, and the Common Issues table provides error-recovery mapping. It falls short of the score-5 anchor because there is no explicit feedback loop ("if validation fails, fix and re-validate") and the destructive `databricks bundle destroy --auto-approve` command appears with no preceding verification step; it is above score 3, which requires missing or merely implicit checkpoints. | 4 / 5 |
Progressive Disclosure | The body is a clear overview that routes detail to four existing one-level-deep reference files via a use-case table ("Configure task types... → references/task-types.md", "Complete worked examples → references/examples.md"), and the task-type and trigger-type tables deep-link to named anchors in those files. All four referenced files exist in the bundle, references are clearly signaled, and detail is appropriately split — an exact match to the score-5 anchor. | 5 / 5 |
Total | 18 / 20 Passed |