{
  "markdown": "# debate\n\nStructured multi-round debate between two AI tools - one proposes, one challenges, and an orchestrator delivers a verdict.\n\n## Why\n\nA single AI tool gives you one perspective. When the stakes are high - architecture decisions, schema design, migration strategies - you want adversarial pressure-testing. debate pits two AI tools against each other in a structured format with enforced rules: the challenger must find flaws before agreeing, both sides must cite evidence, and the orchestrator must pick a winner.\n\nUse cases:\n\n- Stress-test an architecture decision before committing (microservices vs monolith, event sourcing vs CRUD)\n- Get adversarial review of a proposed schema or API design\n- Compare two approaches with structured arguments rather than vibes\n- Force a devil's advocate perspective on a plan you are leaning toward\n\n## Installation\n\n```bash\nagentsys install debate\n```\n\nRequires at least two AI CLI tools installed. See [consult](https://github.com/agent-sh/consult) for individual tool install commands.\n\n## Quick Start\n\n```bash\n# Codex argues for, Gemini challenges - 2 rounds about microservices\n/debate codex vs gemini about microservices vs monolith\n\n# Thorough 3-round debate with explicit flags\n/debate \"Should we use event sourcing?\" --tools=claude,gemini --rounds=3 --effort=high\n\n# Include the current git diff as context for both tools\n/debate codex vs gemini about this refactoring approach --context=diff\n```\n\n## How It Works\n\n1. **Parse** - extracts topic, tools, rounds, and effort from natural language or flags\n2. **Detect** - scans PATH for installed AI CLI tools; requires at least two\n3. **Resolve** - if parameters are missing, presents an interactive picker (proposer, challenger, effort, rounds, context)\n4. **Debate** - executes rounds sequentially. Each round:\n   - The proposer argues for the topic (round 1) or defends against challenges (round 2+)\n   - The challenger identifies flaws, missing considerations, and alternatives\n   - Both must support claims with evidence; unsupported claims are flagged\n5. **Verdict** - the orchestrator (Opus) reads all exchanges and delivers a structured synthesis: winner, agreements, disagreements, unresolved questions, and an actionable recommendation\n\nThe challenger operates under adversarial rules - it must lead with what is wrong or missing before acknowledging agreements, and cannot agree with unsupported claims.\n\n## Usage\n\n### Natural language\n\n```bash\n/debate codex vs gemini about microservices vs monolith\n/debate with claude and codex about our auth implementation\n/debate thoroughly gemini vs codex about database schema design\n/debate codex vs gemini 3 rounds about event sourcing\n```\n\n### Explicit flags\n\n```bash\n/debate \"Redis vs PostgreSQL for caching\" --tools=codex,opencode\n/debate \"Should we use event sourcing?\" --tools=claude,gemini --rounds=3 --effort=high\n/debate \"Performance optimization\" --tools=codex,gemini --context=diff\n```\n\n### Flags\n\n| Flag | Values | Default | Description |\n|------|--------|---------|-------------|\n| `--tools` | TOOL1,TOOL2 | interactive | Proposer and challenger (comma-separated) |\n| `--rounds` | 1-5 | 2 | Number of back-and-forth exchanges |\n| `--effort` | low, medium, high, max | interactive | Reasoning depth for both tools |\n| `--context` | diff, file=PATH, none | none | Codebase context shared with both tools |\n| `--model-proposer` | any model identifier | from effort | Override proposer model |\n| `--model-challenger` | any model identifier | from effort | Override challenger model |\n\n### Debate output format\n\nEach debate produces a structured result:\n\n- **Rounds** - proposer opening, challenger response, and subsequent exchanges displayed as they complete\n- **Verdict** - winner with cited evidence from the debate\n- **Debate quality** - rated on genuine disagreement, evidence quality, and challenge depth\n- **Agreements** - points both sides converged on, with supporting evidence\n- **Disagreements** - points where they diverge, with each side's position\n- **Unresolved questions** - gaps neither side addressed\n- **Recommendation** - actionable next step (the orchestrator must pick a direction)\n\n## Requirements\n\n- [agentsys](https://github.com/agent-sh/agentsys) runtime\n- Node.js (for ACP transport detection)\n- At least two supported AI CLI tools on PATH (claude, gemini, codex, opencode, copilot, or kiro)\n\n## Related Plugins\n\n- [consult](https://github.com/agent-sh/consult) - single-tool consultation for quick second opinions\n- [agentsys](https://github.com/agent-sh/agentsys) - plugin runtime and orchestration\n\n## License\n\nMIT\n",
  "bytes": 4610,
  "sha": "a2b22b601ae0646979d551a88860b45ebeed1598eac1b1400470a98904aad2ce",
  "repo_slug": "agent-sh/debate",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_agent_sh_debate_debate_2ffe0a07/readme"
}