Content
82%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 highly actionable, well-structured entry-point skill: copy-paste-ready script commands, a security checkpoint gate, explicit destructive-op confirmation, and verified one-level-deep bundle references. The main deductions are redundant routing sections and inline blocks (Common Mistakes, SQL Views) that could be moved to references, plus a missing post-load verification step for batch pushes.
Suggestions
Collapse the "Quick Decision Tree" and "Choose the Right Path" table into a single routing block (and drop the duplicate Bundled References re-listing) to remove the body's main redundancy.
Add a verification step after push-data runs (e.g., query the dataset via getData to confirm expected row count, with a retry/escalation note on mismatch) to give the batch flow an explicit feedback loop.
Move the Common Mistakes section (and optionally SQL Views) into a references file such as references/common-mistakes.md, keeping only the top one or two pitfalls inline in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with no concept over-explanation (nothing like 'what a database is'), but there is noticeable redundancy: the Quick Decision Tree and the "Choose the Right Path" table route the same information twice, the Bundled References section re-lists files already linked inline, and the Avoid bullets restate earlier security points. Efficient overall with trim opportunities, matching 4 rather than 3 because each redundancy is small and no token is spent teaching known concepts. | 4 / 5 |
Actionability | Fully executable guidance throughout: copy-paste-ready script invocations with complete flags (--list-accounts, --account-id, --provider/--host/--port/--database/--user/--password-env, push-data with --file/--format/--mode/--confirm-replace), a concrete JSON payload for account search, and per-mistake symptom/why/fix blocks. Not 4 because commands cover the common cases end-to-end rather than leaving minor gaps. | 5 / 5 |
Workflow Clarity | The connect flow is a clear numbered 3-step sequence (account → getDataprovider → createDataprovider), gated by an explicit security checklist ("If ANY checkbox is unchecked, STOP"), and the destructive replace mode requires --confirm-replace yes. It caps at 4 rather than 5 because the batch push flow has no post-load verification step or error-recovery loop (e.g., confirming loaded row counts, or a fix-and-retry path after a failed batch) — though confirmation and mistake-fix guidance keep it above 3's missing-checkpoint level. | 4 / 5 |
Progressive Disclosure | Bundle structure is real and one level deep (references/data-modeling.md, references/connection-types.md, scripts/connect-datasource.js, scripts/push-data.js all exist), and the decision tree plus Bundled References section signal them clearly with one-line descriptions. It falls short of 5 because sizable content blocks — the ~60-line Common Mistakes section and the SQL Views section — live inline in SKILL.md when they would fit naturally in a reference file; this matches anchor 4's "most content appropriately placed, minor organization gaps". | 4 / 5 |
Total | 17 / 20 Passed |