Content
93%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 overview-style skill: token-lean routing tables replace prose, red-flag heuristics are concrete and diff-recognizable, and the bundle cleanly separates overview from detail with condition-gated, one-level-deep references. The single weakness is that the review workflow itself is delegated rather than sequenced, leaving a minor gap in explicit step ordering.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~60 lines of dense routing tables and heuristics with zero explanation of concepts Claude already knows — no Java tutorials, no Spring Boot primers. Every section instructs ("Read only the references that match the changed files", "Formatting, import order, naming style that a formatter or linter enforces"), and even the meta "No duplication" paragraph is brief and load-bearing. This matches the lean, every-token-earns-its-place anchor; there is nothing substantive to trim, so it is not score 4. | 5 / 5 |
Actionability | Concrete, executable guidance throughout: the What-to-Load table maps exact change types to exact files, the severity section defines each level's merge criteria, and the Quick Reference names specific detectable patterns ("Optional.get()", "multi-step writes without @Transactional", "findAll() without paging", "LIKE from user input"). Per the rubric's code-vs-instruction note, absence of code blocks is not penalized when the guidance is this specific — the signals are copy-paste-ready check items covering the common cases, matching the score-5 anchor. | 5 / 5 |
Workflow Clarity | The operative workflow — match changed files, load the corresponding references, flag issues, classify by severity — is unambiguous and driven by explicit routing tables and severity definitions. It is not score 5 because the end-to-end sequence is never laid out as ordered steps (the body deliberately delegates process to the code-reviewer agent), and not score 3 because the flow that does exist is explicit rather than implicit, with no validation gaps relevant to this read-only review task. | 4 / 5 |
Progressive Disclosure | A clear overview that routes to six one-level-deep reference files, all of which exist in the bundle (correctness.md, spring-conventions.md, api-contract.md, security.md, performance.md, smells.md), each gated by a load condition ("Controllers, DTOs, OpenAPI, error handling | references/api-contract.md"). Cross-skill pointers ("conventions → Relational Database Schema (Flyway)") are clearly named, detail is appropriately split into references, and nothing bulk that belongs in a separate file is inlined. This matches the score-5 anchor; it is not score 4 because every reference is well-signaled and navigation is trivial. | 5 / 5 |
Total | 19 / 20 Passed |