Мульти-агентный фреймворк с контролем качества
vibe-coding
multi-agent-ralph-loop is a Claude Code skill that extends Claude Code into an autonomous orchestration framework featuring a 4-layer MemPalace memory stack (L0–L3), six specialized agent teammates, and blocking quality gates. For tasks with complexity ≥ 3, the orchestrator applies Aristotle First Principles decomposition and spawns ralph-coder, ralph-reviewer, ralph-tester, ralph-researcher, ralph-frontend, and ralph-security in parallel via the experimental Agent Teams API, then enforces a 4-stage validation pipeline covering correctness, quality, OWASP-based security (semgrep + gitleaks), and consistency. A continuous learning pipeline extracts decisions and patterns from every session, filters noise—experiments found 46% of auto-learned rules redundant—and graduates only universal patterns to a global Obsidian vault. Core commands include /orchestrator, /parallel, /security, /autoresearch, /adversarial, and /spec, making this framework well-suited for teams building complex systems where security auditing and reproducible architecture decisions are non-negotiable.
- #adr
- #adversarial
- #architecture-diagram
- #ask-questions-if-underspecified
- #audit
- #autoresearch
- #browser-test
- #bugs
- #checkpoint-manager
- #clarify
- #clean-slop
- #code-reviewer