Content
72%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-structured overview skill: correct routing tables, executable CLI workflow, and exemplary progressive disclosure with condition-triggered one-level-deep references. The main gap is workflow validation — operations including deletes and permission changes lack post-operation verification or error-recovery checkpoints in the main body — plus some redundant disambiguation text that could be trimmed.
Suggestions
Add a post-operation verification step to the CLI Workflow (e.g., re-run `get-model-service` after create/update, or confirm grants with the permissions reference's check command) so destructive operations like delete and revoke have an explicit feedback loop.
Consolidate the databricks-model-serving disambiguation, which currently appears three times (description, gateway table, and the 'An explicit request…' paragraph), into the Choose the Correct Gateway section alone.
Merge the Service Types table's per-row 'read references/…' mentions with the 'Querying and permissions' bullet list, which currently repeats the same pointers, into a single navigation block.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence — no basic-concept explanations, and product-specific facts (securables, the legacy `ai-gateway` command-group name) are things Claude would not know. Minor trimming opportunities: the databricks-model-serving disambiguation is repeated in the description, the gateway table, and the "An explicit request…" paragraph, and the Service Types table rows duplicate the "Before any … operation, read references/…" bullets. | 4 / 5 |
Actionability | Concrete, runnable bash commands (`databricks --version`, `databricks ai-gateway --help`, `databricks ai-gateway <operation> --json @payload.json --profile <PROFILE>`) plus exact per-service operation names in the Service Types table. Falls short of 5 because the commands are templated — payload construction and per-operation contracts are deferred to the reference files rather than shown with a worked example. | 4 / 5 |
Workflow Clarity | The CLI Workflow gives a clear numbered sequence with pre-flight checkpoints (verify CLI version, inspect `--help` before building a payload), but there are no post-operation verification steps or error-recovery guidance at the SKILL.md level. Since the operation set includes destructive actions (`delete-model-service`, `delete-mcp-service`, grant/revokes), the missing validation feedback loop caps this at 3. | 3 / 5 |
Progressive Disclosure | Verified against the actual bundle: all 8 files in references/ are cited from the body with conditional "Before X, read references/Y.md" signals keyed to the specific task (querying, permissions, budgets, account-scoped ops, migration), no reference file points to another reference (one level deep), and bulk detail is correctly kept out of the overview. External authoritative-doc links are also clearly grouped. | 5 / 5 |
Total | 16 / 20 Passed |