Content
78%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 well-structured, largely actionable refactoring workflow with explicit verification commands, quantified heuristics, and sensible guardrails (ask-first list, never-do list). Its weaknesses are mild: some philosophy restating knowledge Claude already has, no failure-handling loop in VERIFY, and referenced artifacts (journal format, mapping boundaries) described but never exemplified.
Suggestions
Trim or compress the Philosophy section into one line of team-specific principles — SRP and composition-over-inheritance are knowledge Claude already has and cost tokens without adding guidance.
Add a brief error-recovery loop to VERIFY, e.g., 'If tests or lint fail: fix, re-run, and only proceed when green', to earn a full feedback cycle.
Include one compact example each of a mapping boundary (e.g., a two-line fun UserDto.toUiModel(): UserUiModel) and a properly formatted journal entry, since both are mandated but currently only described.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient with concrete commands and thresholds, but the Philosophy section restates SRP and composition-over-inheritance that Claude already knows, and decorative emoji headers add minor padding — anchor 4, not 5. | 4 / 5 |
Actionability | Gives executable gradle commands ('./gradlew ktfmtFormat', 'testDebugUnitTest', 'lintDebug', 'assembleDebug'), quantified thresholds (500-line ViewModels, 30+ method interfaces, ~300-line change cap) and a worked PR title. Minor gaps: no example of a mapping boundary (toDomain()/toUiModel()) or journal entry despite requiring both. | 4 / 5 |
Workflow Clarity | Clear 5-step sequence (OBSERVE, SELECT & PROPOSE with an approval gate, RENOVATE, VERIFY, PRESENT) with an explicit verification step running four checks. However there is no error-recovery loop (what to do when tests or lint fail), so it fits anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | The ~49-line body has clean, well-labeled sections (Goal, Constraints, Instructions, Examples) with no need for external reference files, and none are bundled — per the rubric's simple-skill guidance this scores 5. | 5 / 5 |
Total | 17 / 20 Passed |