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 high-quality, trap-oriented driver reference: fully executable examples, explicit wrong/right contrasts, and a clean overview-to-references split. Weaknesses are modest — a broken markdown table (the IResultSummary profiling paragraph is inserted mid-table, orphaning the SingleAsync row at line 289), deliberate duplication between the Common Mistakes table and earlier sections, and a reference-style rather than sequenced workflow.
Suggestions
Fix the broken "Cursor consumption methods" table: move the IResultSummary profiling paragraph (SKILL.md line 288) below the table and restore the orphaned `| SingleAsync() | ✅ exactly 1 | ❌ | Expect one row |` row (line 289) as a proper table row.
Reduce duplication between the Common Mistakes table and preceding sections — keep only rows not already illustrated inline (e.g., await using, cursor-after-loop, and the AsObject using directive are each covered twice) — and consolidate version-gated notes (the 6.4.0 pin, v6.3.0 profile changes, 6.2 UUID support) into a single version-compatibility section.
Close the batch-operation feedback loop: after the UNWIND example, verify the write by checking `summary.Counters.NodesCreated` against the input count before declaring success.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Largely lean — terse rules, tables, and code with no explanation of concepts Claude already knows — but the 22-row Common Mistakes table substantially restates rules already covered in preceding sections (await using, cursor rules, AsObject using directive, rollback isolation), and time-sensitive version notes sit inline ("Pin Neo4j.Driver >= 6.4.0", "[v6.3.0+]", "[driver 6.2+, Bolt 6.1, Neo4j 2026.08+]") rather than in a dedicated version/deprecated section. Efficient with minor trimmable redundancy — anchor 4, not 5; not 3 because there is no padding or generic filler. | 4 / 5 |
Actionability | Every section gives complete, executable C#: `dotnet add package Neo4j.Driver`, full DI registration including a hosted shutdown service, fluent ExecutableQuery examples with parameters/config, managed-transaction callbacks, UNWIND batching, object mapping, and typed catch blocks with an `ex.Code` filter. Copy-paste ready and covering the common cases — anchor 5; not 4 because there is no pseudocode and no missing key details. | 5 / 5 |
Workflow Clarity | Strong checkpoints are present — the "Choose the Right API" decision table, ❌/✅ contrast patterns, `VerifyConnectivityAsync()` fail-fast, counters checks, rollback isolation, idempotent-write guidance for commit uncertainty, and a final Checklist — but the document is a reference layout rather than a sequenced task flow, and the UNWIND batching example lacks an inline post-write verification step. Anchor 4; not 5 because there is no explicit end-to-end sequence, and not 3 because validation/verification is substantially present, so the database/batch cap does not bind. | 4 / 5 |
Progressive Disclosure | The body carries the high-frequency core patterns while genuinely advanced material is split into three real, one-level-deep files, each clearly signaled under "Load on demand" with a one-line description: references/transactions.md (bookmarks, TransactionConfig, commit uncertainty), references/performance.md (spatial types, pool tuning, streaming), references/object-mapping.md (blueprint/lambda mapping, repository pattern) — all verified present in the bundle. Matches anchor 5's overview-plus-labeled-references structure; not 4 because nothing that plainly belongs in a reference is inlined and navigation is easy. | 5 / 5 |
Total | 18 / 20 Passed |