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 high-quality, dense reference body: executable code for every common case, a decision table for API selection, explicit verification checkpoints, and a well-wired one-level-deep reference bundle. Weaknesses are minor: some duplicated content between body and references, a redundant checklist, and no explicit retry loop for batch writes.
Suggestions
Remove the duplicated record.data() mapping prose in 'Result Access & Null Safety' (keep one of the two back-to-back statements) and either drop the closing Checklist or the overlapping rows of the Common Errors table to reclaim tokens.
Add a short validate-retry note to 'Batch Writes with UNWIND': verify summary.counters against len(rows) and split/retry failed chunks, giving the batch path an explicit feedback loop.
De-duplicate content shared with references (e.g., keep only a pointer or a minimal inline snippet for the FastAPI lifespan pattern and pool tuning kwargs) so SKILL.md reads as a lean overview over the reference files.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and driver-specific throughout — version-gated facts ("Pin `neo4j>=6.4.1`"), the trailing-underscore convention, Result lifetime rules — with no padding explaining concepts Claude already knows. It is not level 5 because some content is duplicated: the `record.data()` mapping (Node → dict, Relationship → tuple, Path → list) is stated twice in back-to-back lines in "Result Access & Null Safety", and the closing Checklist largely restates the Common Errors table. It is above level 3 since the redundancy is minor trimming, not unnecessary explanation. | 4 / 5 |
Actionability | Every section ships copy-paste-ready executable code covering the common cases: driver creation with URI schemes and auth options, `execute_query` with `routing_`/`database_`, managed transaction callbacks, async singleton + FastAPI lifespan + `asyncio.gather`, exception handling with correct catch order, and pool tuning with concrete kwargs and defaults. This matches the anchor for fully executable, copy-paste-ready code covering the common cases. | 5 / 5 |
Workflow Clarity | A clear selection sequence is present (Installation → Environment Variables → Driver Lifecycle → "Choosing the Right API" decision table → per-API sections → Error Handling → Common Errors → Checklist) with explicit verification checkpoints: `verify_connectivity()` at startup, checking `summary.counters.nodes_created` after writes, `.consume()` to guarantee commit before proceeding, and `single(strict=True)` to validate exactly-one-row results. It is not level 5 because UNWIND batch writes lack an explicit validate→fix→retry loop (e.g., verifying counters against the row count and retrying failed chunks), but checkpoints are present so it stays above level 3. | 4 / 5 |
Progressive Disclosure | All four referenced files (async.md, data-types.md, performance.md, transactions.md) exist, are exactly one level deep with no cross-nesting, and are signaled inline at the relevant sections ("Full async patterns → [references/async.md]") plus a References index with per-file descriptions. It is not level 5 because some content is duplicated between the body and the references (the FastAPI lifespan pattern also appears in references/async.md; pool kwargs also appear in references/performance.md) and the ~420-line body is denser than a lean overview, which are minor organization gaps per the level-4 anchor. | 4 / 5 |
Total | 17 / 20 Passed |