Content
56%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 thorough, code-heavy reference that is highly actionable across the full Zarr feature surface, held back by noticeable verbosity (repeated chunk-size/metadata guidance across four-plus sections) and a progressive-disclosure defect: the bundled references/api_reference.md is never linked from SKILL.md, so it is effectively undiscoverable. Trimming duplication and splitting patterns/API detail into the existing reference file would materially improve it.
Suggestions
Link references/api_reference.md from the body (e.g., '## API reference — see [api_reference.md](references/api_reference.md)') so the bundled file is actually discoverable, and move the compression codec details and 'Common Patterns' walkthroughs there.
Deduplicate the chunk-size guidance (currently repeated in Chunking Strategies, the Performance checklist, the Cloud-Native pattern, and Common Issues) into a single canonical section.
Fix minor executability gaps: define `data`/`data2`, import pandas in the Xarray example, and separate or annotate v2 vs v3 API usage (e.g., S3Map + create_array).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose at ~770 lines: chunk-size guidance is repeated in four places (Chunking Strategies, the Performance Optimization checklist, the Cloud-Native pattern, and Common Issues), consolidated-metadata advice appears in both 'Cloud Storage Best Practices' and its own section, and there are several padded 'Benefits' bullet lists and explanatory asides (e.g., 'Groups organize multiple arrays hierarchically, similar to directories or HDF5 groups') that restate knowledge Claude already has. It is not a score-1 tutorial-for-beginners dump — most content is code, not prose — but the duplication and padding clearly sit below the 'mostly efficient' score-3 anchor. | 2 / 5 |
Actionability | Nearly all guidance is concrete, executable code covering creation, indexing, chunking, compression, storage backends, cloud I/O, groups, Dask/Xarray integration, and diagnosis of common issues (e.g., 'print(z.chunks)' followed by specific solutions). Minor gaps keep it from a 5: some examples use undefined variables (`data` in the Cloud-Native pattern, `data2` and an unimported `pd` in the Xarray example), and the API mixes v2-style (`s3fs.S3Map`, `zarr.append`) with v3-style (`zarr.create_array`, `shards=`) calls that may not run together as written. | 4 / 5 |
Workflow Clarity | For a library/reference skill, the body provides a clear decision sequence — a numbered 'Checklist for Optimal Performance' (chunk size → chunk shape → compression → backend → sharding → parallel I/O) plus 'Common Issues' sections structured as diagnosis code → concrete solutions, which function as feedback checkpoints. It falls short of 5 because there is no explicit end-to-end workflow with validation/error-recovery loops, and a few sequences (e.g., the Cloud-Native pattern's write-then-consolidate flow) omit checkpoints like verifying the write or handling partial failures. | 4 / 5 |
Progressive Disclosure | Section headers are well-organized and one-level-deep, but the bundle contains references/api_reference.md (a 12KB function reference) that is never mentioned or linked anywhere in SKILL.md — an agent following the skill would never discover it. Meanwhile large blocks of inline content (compression codec tables, the four 'Pattern' walkthroughs, profiling details) belong in reference files per the score-3/4 anchors. This matches 'references present but not clearly signaled; content that should be separate is inline'; it is above score 2 because the document is well structured, not a minimal-structure wall of text. | 3 / 5 |
Total | 13 / 20 Passed |