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 exemplary decision-focused skill body: terse, well-sectioned, and anchored by a concrete mapping-table template and a gotchas list of real failure modes. The only gaps are a slightly vague fallback branch in the mechanism decision and the absence of any executable code or a post-table validation step.
Suggestions
Tighten the third mechanism branch into a concrete check (e.g. 'if you find yourself writing a per-message loop or building chunks by hand, stop and ask which reader/lens covers it').
Add one short code snippet (or a completed minimal mapping-table row to code correspondence) to anchor the table directly to executable ingest calls.
Add an explicit post-table checkpoint: verify every row has an entity path, archetype, timeline, and static/temporal + base/layer decision before writing conversion code.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense, lean declarative prose with zero padding — every line is a decision, a rule, or a concrete gotcha (e.g. "Static shadows all temporal data of that component for all time"); it assumes Claude's competence throughout. | 5 / 5 |
Actionability | A copy-paste-ready filled mapping table, concrete archetype names, and exact flags like `send_properties=False` make the guidance mostly executable; one branch ("consider if you are about to hand-build something… ask for clarification") stays soft, and no code snippet is given even where a one-liner would anchor the pattern. | 4 / 5 |
Workflow Clarity | The sequence (pick mechanism → model → decisions → produce the mapping table before coding) is clear and the table is an explicit reviewable checkpoint; not a destructive/batch skill, but there is no validate-and-fix loop after the table is filled in, keeping it below a 5. | 4 / 5 |
Progressive Disclosure | Self-contained with no bundle files needed; sections are well-organized and every external pointer (rerun-chunk-processing, the importer skills, rerun-catalog-queries, rerun.io docs) is clearly signaled one level deep. | 5 / 5 |
Total | 18 / 20 Passed |