ry-rules-review

name: ry-rules-review description: "Аудит реализации против rldyour rules. Используй для: /ry-rules-review, проверь по правилам, аудит правил, проверь жесткие правила, качество по правилам. EN triggers: rules review, hard rules audit, check against rules, rldyour rules check, policy compliance audit, rules audit."

Review a diff, PR, branch, file scope, or implementation against rldyour rules.

Determine target: current diff, branch vs main, PR, file scope, or user-provided scope.
Use Serena-first code inspection for affected symbols and integration paths (get_symbols_overview, find_symbol, find_referencing_symbols).
Apply rule skills:
- quality-first-engineering - hard bans, semantic entropy, scope policy.
- architecture-boundaries - FSD/VSA/Hexagonal placement, ADR triggers.
- implementation-discipline - integration sync, reuse, error handling.
- dependency-compatibility-policy - compatibility, supply chain, lockfile discipline.
- verification-quality-gates - required checks per change type.
- project-instructions-policy - AGENTS.md/.claude/CLAUDE.md/REVIEW.md/ADR durability.
Use tech-research or web-research skills when the review depends on current technology behavior, dependency versions, or architecture best practices.
Report findings in Russian, ordered by severity and confidence.
Default mode is report-only. Modify files only if the user explicitly asks to fix findings.

Severity: critical, high, medium, low.
Confidence: 0-100.
Location: exact file and line when possible.
Rule: which rldyour rule is violated.
Evidence: concrete code, config, test, or docs evidence.
Impact: what becomes incorrect, fragile, unscalable, insecure, or harder to maintain.
Fix: actionable correction.
Disposition: must-fix, should-fix, ask-user, or defer.

Drop confidence <30. Validate confidence 30-49 against extra evidence before reporting.

Read references/rules-policy.md first; load other references only when they match the reviewed issue.