ry-review

name: ry-review description: "Глубокое ревью diff/PR/scope с research и параллельными reviewer tracks. Report-only по умолчанию. Используй для: проверь реализацию, сделай ревью, найди проблемы, инспекция кода, audit diff. EN triggers: ry-review, code review, audit changes, deep review, multi-track review, report-only review, reviewer subagents."

Find real issues before merge or deploy. Default mode is report-only: do not edit files unless the user explicitly asks after seeing findings.

Determine review target: current diff, branch vs main, PR, file scope, or prompt scope.
Initialize missing context with ry-init if needed.
Use Serena to map changed symbols and affected integration graph. Prefer serena_find_symbol, serena_get_symbols_overview, and serena_find_referencing_symbols before raw file reads.
Use ry-explore for current implementation best practices when the review depends on external technology behavior. Prefer Context7 MCP for official versioned docs, DeepWiki MCP for public repo architecture, and Grep MCP for real production usage patterns.
Run reviewer tracks. Use subagents when this review request or an explicit-review ry-start request calls for parallel review. Reviewer subagents are defined in .opencode/agents/ with mode: subagent, hidden: true, permission: { edit: "deny" }.
Consolidate findings by severity and confidence. Validate uncertain findings with code evidence.
Output Russian report with exact paths, impact, suggested fixes, and whether each finding is must-fix.

When the owner asks for a time-window or history-based review, resolve the review target before dispatching reviewers:

last N days / за последние N дней: compute the exact commit range with git log --since.
since DATE / с DATE: use the explicit date as the lower bound.
PR #N or issue #N: inspect the GitHub PR/issue and verify its relevance against current code before treating it as a finding.
branch vs main: review the merge-base diff plus related commits.
since last deploy: identify the last verified deployment marker or ask for it if the repository has no durable deploy record.

Report the resolved commit range, PR/issue IDs, and any unresolved evidence gap.

Read references/reviewer-protocol.md. These tracks are orchestrated by ry-review or ry-start; they are not broad implicit-entry skills.

Architecture review: module boundaries, layer violations, coupling, abstraction quality.
Quality review: code smells, duplication, dead code, error handling, naming, readability.
Consistency review: pattern alignment, naming conventions, import structure, style uniformity.
Integration review: API contracts, data flow, cross-module interactions, breaking changes.
Verification review: test coverage, edge cases, regression risk, observable behavior.
Security review (flow-security-review): when scope is sensitive or explicitly requested.

Each track produces findings with:

Path: exact file and line reference.
Severity: must-fix / should-fix / nit / observation.
Confidence: high / medium / low - low-confidence findings must include the evidence gap.
Description: what is wrong and why it matters.
Suggested fix: concrete remediation or next step.

Report in Russian: