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.
A tight, well-organized instruction skill: domain-specific knowledge only, a concrete tool-selection table, and a clear search workflow with fallbacks. The main gaps are semi-placeholder tool signatures instead of fully executable call examples, and error-recovery guidance that is separated from the drill-down sequence rather than integrated as checkpoints.
Suggestions
Show one fully literal example call chain (e.g. find → read .overview.md → scoped grep → read with offset/limit) so the signatures in the table become copy-paste ready.
Fold the no-hits and low-score recovery guidance into the Drill down steps as explicit checkpoints (e.g. step 2b: 'if all scores < 0.4, rephrase with the domain's terms') instead of deferring it to Tips.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout — the .abstract.md/.overview.md convention, the tool-selection table, the "Low scores (under ~0.4) are usually noise" threshold — all OpenViking-specific knowledge Claude does not already have, with no padding or explanation of known concepts. Matches anchor 5: every token earns its place. | 5 / 5 |
Actionability | Concrete invocation patterns are given ("find(query, target_uri=\"viking://resources\")", "glob(pattern, uri) e.g. **/*.pdf", "read(uris=[...])") plus a numbered drill-down, but several entries are signature placeholders rather than fully copy-paste-ready calls — mostly executable with minor gaps, matching anchor 4 rather than anchor 5. | 4 / 5 |
Workflow Clarity | The drill-down gives a clear 4-step sequence with a stopping rule ("Stop as soon as you have what the task needs") and feedback loops (no-hits → "list viking://resources"; low scores → rephrase), but the recovery guidance lives in a separate Tips section rather than as explicit checkpoints inside the sequence — fits anchor 4 (clear sequence, minor validation gaps). No destructive/batch cap applies since this is a read-only search skill. | 4 / 5 |
Progressive Disclosure | The skill is under 50 lines with no bundle files and no need for external references; the body is cleanly organized into "Pick the tool", "Drill down", and "Tips" sections with no dangling file references. Per the rubric's simple-skill guideline, well-organized sections alone merit anchor 5. | 5 / 5 |
Total | 18 / 20 Passed |