DGouron

Follow-up review that never trusts a commit message — always re-reads the actual diff before resolving a thread. Cites a real source for new issues, matching review-advanced.

Rigorous sequential-audit code review with a dedicated security block and cited pedagogical lessons. Customize for your project.

Review de suivi qui ne fait jamais confiance à un message de commit — relit toujours le diff réel avant de résoudre un thread. Cite une source réelle pour les nouveaux problèmes, comme review-advanced.

Code review rigoureuse à audits séquentiels, avec bloc sécurité dédié et leçons pédagogiques sourcées. À personnaliser pour votre projet.

Complete review of a documentation-focused MR/PR with 5 sequential audits oriented for documentation projects (markdown-quality, link-validity, terminology, freshness, examples-validity). No code-architecture audits, no React patterns, no SOLID. An orchestrator runs each audit one by one. Generates an .md report and posts it directly on the MR/PR. Direct mode with sourced lessons.

Follow-up review to verify corrections on a MR. Sequential execution to avoid memory spikes. Checks blocking issues, detects new problems, and posts a concise report on GitLab.

Follow-up review to verify fixes. Uses context file for thread management.

Basic code review skill template. Customize for your project.

Audit naming quality (files, functions, types, variables, folders) without reading the code. Takes names as input and reports what Claude understands from each name alone. Judges three axes: Screaming Architecture (does the name reveal the domain?), Clean Code (full words, intention-revealing, no abbreviations), and Ubiquitous Language (vocabulary consistency across the codebase). Useful to validate that names self-document.

Extracts business rules from a module (backend + frontend) and produces three tables: Product view (business concepts), Backend dev view (type + source), Frontend dev view (type + source). Saves to docs/business-rules/.

Strategic DDD guide for this project (fullstack — backend + frontend share the same bounded contexts). Use to slice the domain into bounded contexts, define ubiquitous language, create a new business module, analyze boundaries between contexts. Tactical patterns follow Clean Architecture (see architecture-backend and architecture-frontend skills).

Runs a Big Picture Event Storming session on a bounded context or the entire codebase (backend + frontend). Discovers domain events, commands, entities, context boundaries, frontend projections (hooks, presenters, views). Produces a structured document in docs/ddd/.

Autonomous frontend feature implementation via spec-driven development. Orchestrates a frontend-planner and a frontend-implementer with architecture-frontend preloaded. Applies Humble Object pattern to views, MVVM structure with Presenter classes + Zod ViewModels, singleton registry DI. Consumes specs produced by /product-manager. For backend features, use /implement-feature instead.

Feature challenge and specification. Use to define a feature, write acceptance criteria, scope a ticket, produce INVEST specs with custom DSL in docs/specs/. Refuses vague scopes and forces clarification.

Discover implicit team conventions (tribal knowledge) in a codebase. Spawns parallel sub-agents to analyze 6 axes: naming, error handling, testing, architecture, git workflow, and code idioms. Generates a structured report in docs/tribal-knowledge/.

Clean Architecture guide for the React frontend. Use to create frontend modules, entities, use cases, presenters, hooks, views, gateways, guards. Applies Humble Object pattern (Uncle Bob) — views are untestable glue, hooks and presenters hold all logic. For backend architecture, use architecture-backend instead.