Content
57%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.
The body delivers genuinely concrete, well-parameterized Neuropixels pipeline guidance with a correct SpikeInterface-based example, but it is padded (Key Features duplicates the other sections), the example is not actually runnable as shipped (imports an unprovided 'neuropixels_analysis' module), validation checkpoints are missing for a batch pipeline, and its navigation to the real bundle is broken — wrong reference directory name, unlisted ANALYSIS.md, and completely unmentioned scripts/ and assets/ contents.
Suggestions
Fix the reference paths from 'reference/...' to 'references/...', render them as markdown links with a one-line note on when to consult each (e.g. 'PREPROCESSING.md — full filter/bad-channel/CAR chain and parameters'), and add a section listing 'scripts/' (notably neuropixels_pipeline.py, run_sorting.py, compute_metrics.py, export_to_phy.py) and 'assets/analysis_template.py' so the whole bundle is discoverable from SKILL.md.
Delete the 'Key Features' section (it restates 'When to Use' and 'Implementation Details') and trim known-concept explanations and hedging qualifiers ('implementation-dependent', 'dataset-dependent; document your choices') to cut roughly a quarter of the body's tokens.
Add explicit validation checkpoints to the workflow: inspect the drift map before deciding on correction, sanity-check unit counts and QC-metric distributions before applying curation thresholds, and verify the Phy and NWB exports completed (e.g. check output folder contents) before declaring the run finished.
Either document how to obtain/install the 'neuropixels_analysis' package (it is absent from the bundle and the Dependencies list) or rewrite the example's npa.* calls to call the provided scripts/ functions or direct SpikeInterface equivalents (si.estimate_motion, si.correct_motion) so the script is runnable as written.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient technical content (parameters, code, pipeline steps), but the 'Key Features' section (~24 lines) substantially duplicates 'When to Use' and 'Implementation Details', and there are several explanations Claude already knows ('Common reference (often median CAR) to reduce shared noise', 'low ISI violations (refractory period respected)', 'Motion/drift can strongly degrade sorting quality') plus hedging qualifiers ('implementation-dependent', 'dataset-dependent; document your choices') that add tokens without information. This is 'some unnecessary explanation or could be tightened' rather than the noticeably-verbose anchor 2, since the majority of the body is dense, actionable guidance. | 3 / 5 |
Actionability | The example script uses correct, concrete SpikeInterface API calls (read_spikeglx, run_sorter, create_sorting_analyzer, the analyzer.compute chain, export_to_phy) with specific parameters, and Implementation Details gives tunable values (freq_min 300-400 Hz, batch_size, Th_learned, QC thresholds). However, it is not the claimed 'complete, runnable script': it imports 'neuropixels_analysis as npa' — a module absent from the bundle and missing from the Dependencies section — so 4 of the 8 steps (preprocess, estimate_motion, curate, report) fail at import, and the bundle's own implementing scripts (scripts/neuropixels_pipeline.py with preprocess/curate_units/correct_motion) are never referenced. Concrete code with a non-minor gap fits anchor 4 more than anchor 3 (real executable code, not pseudocode) but not anchor 5. | 4 / 5 |
Workflow Clarity | The sequence is clear (numbered steps 1-8 in the example, mirrored by Implementation Details sections 1-7) and there is one explicit conditional checkpoint ('if max drift exceeds 10 µm, apply correction'). But this is a long-running batch pipeline and validation/verification steps are otherwise absent — no instruction to inspect the drift map before deciding, sanity-check unit counts or QC distributions before curation, or verify the Phy/NWB exports. Per the rubric's cap, a batch workflow without validation checkpoints cannot score above 3; it sits above anchor 2 because the sequence itself is coherent and one checkpoint does exist. | 3 / 5 |
Progressive Disclosure | Section structure is good and a reference list is present, but every one of the 9 listed reference paths is wrong ('reference/standard_workflow.md' etc. while the actual bundle directory is 'references/'), the list is buried at the end of 'Key Features' as plain paths with no links or per-file descriptions, 'references/ANALYSIS.md' is omitted, and the entire 'scripts/' bundle (6 pipeline scripts) and 'assets/analysis_template.py' are never mentioned — orphaned content. Meanwhile Implementation Details sections duplicate what the dedicated reference files (PREPROCESSING.md, MOTION_CORRECTION.md, etc.) presumably cover, i.e. content that belongs in separate files is inlined. This is 'some structure, references present but not clearly signaled' (anchor 3) rather than anchor 4, since the navigation to the bundle is effectively non-functional as written. | 3 / 5 |
Total | 13 / 20 Passed |