Content
75%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 strong, highly actionable body: fully executable code for every pipeline stage, a clearly sequenced workflow with real validation checkpoints, and a well-verified one-level reference structure. The main weakness is token efficiency — scattered version/date pins and redundant script documentation could be consolidated or pushed to references.
Suggestions
Consolidate version- and date-sensitive notes (SpikeInterface 0.105.0 behaviors, the read_phy bug, zarr pin, Python 3.13 test env) into a single 'Version compatibility' or 'Known issues' section instead of scattering them across workflow steps.
Trim the 'Bundled Resources' section: the per-script one-liners duplicate the Quick Start pipeline call and the inline references to each script elsewhere; a compact table of script → purpose with pointers to the reference guides would save substantial tokens.
Move 'Key Parameters to Adjust' and 'Project Structure' into the existing references (e.g. standard_workflow.md or api_reference.md) and link to them, keeping SKILL.md a leaner overview of the workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operational with no conceptual over-explanation, but version/date pins are scattered across many sections ('Targets SpikeInterface 0.105.0... (reviewed 2026-10-01)', 'These flags match SpikeInterface 0.105.0', 'SpikeInterface 0.105.0 has no export_to_nwb', 'zarr>=2.18,<3', Python 3.13) rather than consolidated, and 'Bundled Resources' re-documents scripts already shown in Quick Start. This fits 'mostly efficient but could be tightened' rather than 'minor instances', especially given the rubric's explicit penalty for scattered time-sensitive information. | 3 / 5 |
Actionability | Every workflow step ships complete, executable code with real API calls and arguments (read_spikeglx with stream_name, run_sorter with do_correction=False, a concrete pandas curation query, export commands), plus runnable bash commands for all bundled scripts and specific parameter ranges. Code is copy-paste ready and covers the common cases (SpikeGLX/Open Ephys/NWB, multiple sorters). | 5 / 5 |
Workflow Clarity | A clear numbered 1–8 sequence (preprocess → drift → sort → postprocess → curate → export) with genuine validation checkpoints ('Validate AP stream, calibration, channel order, probe geometry and segment boundaries first', 'Always inspect drift before sorting', 'validate the resulting NWB file', inspect model requirements before loading). It stops short of the top anchor because explicit error-recovery feedback loops (validate → fix → retry) are not spelled out. | 4 / 5 |
Progressive Disclosure | Good structure: the core workflow is inline, and all ten detailed guides live in references/ (verified to exist, one level deep, clearly presented in a table plus in-context links), with scripts and a template bundled. However, the ~450-line body also carries the full workflow, key parameters, pitfalls, project structure, and a citation block — heavier than a lean overview with some content that belongs in references, matching 'good structure with minor organization gaps'. | 4 / 5 |
Total | 16 / 20 Passed |