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.
An unusually dense, high-signal skill body: executable code, non-obvious operational facts, and explicit validation checkpoints. Its main weakness is structure — everything lives in one monolithic file with no reference bundle, where advanced topics should be split out.
Suggestions
Split advanced topics (TypeSpec, PDE discovery, dimensional constraints, complexity_mapping, parallelism/Slurm) into reference files under references/ and link them from short summary bullets in SKILL.md.
Add explicit 'if this fails, do X' recovery pointers to the Recommended workflow steps, tying each step to the relevant troubleshooting-table row.
Move the deprecation table and version notes into a versioned reference file so SKILL.md stays stable as PySR versions change.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is exceptionally lean: no filler explaining what symbolic regression is or how libraries work, and nearly every sentence carries PySR-specific, non-inferable facts (e.g. the exact oversize-guess warning text, 'JULIA_NUM_THREADS is inappropriate under juliacall', pickles being version-locked). | 5 / 5 |
Actionability | Copy-paste-ready code throughout — quick start, guesses, TemplateExpressionSpec, TypeSpec, custom operators with typed NaN guards, and complete loss_function/complexity_mapping implementations — with concrete parameter values covering the common cases. | 5 / 5 |
Workflow Clarity | The numbered 'Recommended workflow' is clearly sequenced with explicit validation steps (brief debug run, held-out predict, monitoring hall_of_fame.csv, forward simulation from held-out initial conditions), but error-recovery loops are delegated to the troubleshooting table rather than wired into the workflow itself. | 4 / 5 |
Progressive Disclosure | No bundle files exist and this ~360-line monolithic SKILL.md inlines several advanced topics — TypeSpec, PDE discovery, dimensional constraints, complexity_mapping — that clearly belong in separate reference files, though section headers and a front-loaded quick start keep it navigable. | 3 / 5 |
Total | 17 / 20 Passed |