By dynamic-dome
Adversarial agent swarms with Devil's Advocate debate pattern. Spawns Scanner, Advocate, and Fixer agents for quality-driven analysis through structured debate.
Devil's Advocate debate agent. Takes a prosecutor or defender role and argues for/against analysis findings through structured rounds. Produces verdict proposals with accept/reject classifications. Used in pairs for adversarial quality filtering.
Applies minimal, targeted fixes for consensus-accepted findings from adversarial debate. Reads accepted findings, implements the smallest possible change, runs tests, and documents each fix. Skips fixes that would break other code.
Parallel analysis agent in an adversarial swarm. Performs focused scanning of a specific scope (security, code quality, architecture, etc.) and reports all findings with severity, confidence, and evidence. Prefers over-reporting to missing real issues.
Orchestrates adversarial agent swarms: decomposes goals into scanner assignments, manages debate rounds between advocates, synthesizes consensus, and dispatches fixers. Use this agent when a task benefits from structured adversarial analysis (security audits, architecture reviews, code quality checks).
REDIRECT: Use agentic-os:research-pipeline instead. This is an alias to avoid duplication.
Orchestrate adversarial agent swarms with Devil's Advocate debate pattern: Scanner agents analyze in parallel, two Advocates debate findings for up to 3 rounds, a Team Lead synthesizes consensus, and Fixer agents apply the accepted changes. Trigger phrases: "adversarial swarm", "devil's advocate", "adversarial review", "agent debate", "swarm review", "start swarm", "schwarm starten", "teufels advokat", "adversarialer review".
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Adversarielle Agent-Teams fuer Claude Code — bessere Ergebnisse durch strukturierte Debatte.
Statt einen Agenten linear arbeiten zu lassen, spawnt ein "Team Lead" mehrere Sub-Agenten mit unterschiedlichen Rollen. Zwei "Devil's Advocate"-Agenten debattieren die Ergebnisse — erst bei Konsens wird gehandelt.
┌──────────────────────────────────────────────────────┐
│ Team Lead (Orchestrator) │
│ │ │
│ ├── Scanner 1 ──┐ │
│ ├── Scanner 2 ──┤ │
│ ├── Scanner N ──┤ │
│ │ ▼ │
│ │ ┌─────────────┐ │
│ │ │ Scratch Pad │ (geteilte Ergebnisse) │
│ │ └──────┬──────┘ │
│ │ ▼ │
│ │ ┌───────────────────────┐ │
│ │ │ Devil's Advocate 1 │◄──► Debatte │
│ │ │ Devil's Advocate 2 │ │
│ │ └───────────┬───────────┘ │
│ │ ▼ │
│ │ Konsens erreicht? │
│ │ ja │ │ nein │
│ │ ▼ └──► weitere Runde │
│ │ Fixer-Agent │
│ │ │ │
│ │ ▼ │
│ │ Code-Aenderung │
└──────────────────────────────────────────────────────┘
devil-advocate-swarms/
├── CLAUDE.md # Projekt-Instruktionen
├── README.md # Diese Datei
├── .claude-plugin/
│ └── plugin.json # Plugin-Manifest
├── agents/ # Single Source of Truth fuer Agent-Prompts
│ ├── swarm-team-lead.md # Team Lead — Adversarial Swarm Orchestrator
│ ├── swarm-scanner.md # Scanner Agent — Parallele Funde-Erfassung
│ ├── swarm-advocate.md # Devil's Advocate (prosecutor/defender)
│ └── swarm-fixer.md # Fixer Agent — Konsens-Fix-Anwendung
├── commands/
│ └── swarm.md # /swarm Slash-Command
├── scripts/
│ └── orchestrator.sh # Cross-Provider Bash-Pipeline (Codex+Sonnet+Opus)
├── examples/
│ ├── security-audit.md # Beispiel: Security-Audit Workflow
│ └── design-review.md # Beispiel: Architektur-Review Workflow
├── skills/
│ ├── swarm-orchestrator/
│ │ └── SKILL.md # /swarm Skill (Agent-Tool-Modus)
│ └── research-pipeline/
│ └── SKILL.md # Redirect-Alias auf agentic-os:research-pipeline
└── .agent-memory/ # erst beim Lauf gefuellt
├── scratch/ # Geteiltes Scratch Pad
├── consensus/ # Konsens-Ergebnisse
└── debates/ # Debatte-Logs
Hinweis zur Architektur: Bis 2026-04-30 existierte zusaetzlich ein prompts/-Verzeichnis
mit denselben Rollen-Texten in einer aelteren Format-Variante. Es wurde aufgegeben, weil
der scripts/orchestrator.sh die Prompts inline im Bash haelt (siehe DECOMPOSE_PROMPT=,
SCANNER_PROMPT=, etc.) und prompts/ damit faktisch toter Code war.
Single Source of Truth fuer Agent-Bodies ist heute agents/.
// .claude/settings.json
{
"experimental_agent_teams": 1
}
| Use Case | Scanner | Advocate-Thema | Fixer |
|---|---|---|---|
| Security Audit | 5-10 Vulnerability Scanner | "Ist das ein echtes Risiko?" | Patch-Agent |
| Architektur Review | 2-3 Design Analyzer | "Microservice vs. Monolith?" | Refactor-Agent |
| Copywriting | 2 Stil-Varianten | "Minimalistisch vs. Warm?" | Final-Writer |
| Code Quality | 3 Linter/Reviewer | "False Positive oder Bug?" | Bugfix-Agent |
Jeder Subagent = eigenes Kontextfenster. Ein Schwarm von 10 Scannern + 2 Advocates verbraucht ca. 12x so viele Tokens wie ein Einzelagent. Start klein (2-3), dann skalieren.
Phase: Planung — Projektstruktur angelegt, Use Cases definiert. Naechster Schritt: Prompt-Templates fuer Scanner und Advocate Rollen.
npx claudepluginhub dynamic-dome/devil-advocate-swarms --plugin devil-advocate-swarmsAdapter-agnostischer 4-Rollen-Loop (Orchestrator/Researcher/Builder/Judge) mit persistierenden State-Dateien pro Projekt.
Autonomous improvement loop for Claude Code projects — decomposes goals into tasks, executes them via Worker/Judge subagents, mines session logs for error patterns, and generates evidence-based CLAUDE.md improvement suggestions.
Interactive Claude Code workshop plugin — two modes: /workshop guide (moderator co-pilot) and /workshop learn (self-paced tutor). 17 modules (5+5+7) across 3 blocks covering foundations, ecosystem, and advanced topics.
Autonomer Meta-Agent-Orchestrator: Ein Opus 4.6 Agent der Haiku-Brainstormer, NotebookLM-RAG und Codex-Instanzen orchestriert um komplexe Aufgaben autonom zu loesen.
Suite of 7 skills, 7 slash commands, and 5 read-only checker hooks that turns rough ideas into agentic workflow blueprints, runs adversarial review-fix-verify loops, audits ecosystems as dynamic graphs, curates raw inbox material into wiki pages, runs MCP/DCO actions safely, and gates skill promotion. Built from the wiki/patterns/agentic-workflow-blueprints reference.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Complete developer toolkit for Claude Code
Intelligent draw.io diagramming plugin with AI-powered diagram generation, multi-platform embedding (GitHub, Confluence, Azure DevOps, Notion, Teams, Harness), conditional formatting, live data binding, and MCP server integration for programmatic diagram creation and management.
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.