{
  "markdown": "# Agent Discussion\n\nA Claude Code skill that lets AI agents from different platforms (Claude Code, Antigravity, Copilot, Gemini CLI, etc.) hold a structured, autonomous discussion about any topic and reach agreement through a shared file.\n\nNo servers. No MCP. No orchestrator. No setup scripts. Each agent reads one skill file and handles everything — intros, turn-taking, convergence.\n\n## How It Works\n\n1. In Agent A: say `/agent-discussion` → it asks for the topic, creates the discussion file, writes an unbiased intro, then takes its first position\n2. In Agent B: say `/agent-discussion` → it finds the existing file, writes its own unbiased intro, then enters the debate\n3. You watch (or go get coffee)\n\nEach agent starts with an **unbiased overview** of the topic before taking a position, grounding the discussion in shared facts rather than entrenched advocacy.\n\n## Quick Start\n\n```\n# Terminal 1 — Claude Code\n> /agent-discussion\nClaude: What name should I use? → \"claude\"\nClaude: File path? → \"./discussion.md\"\nClaude: Topic? → \"REST vs GraphQL for our new API\"\nClaude: Created discussion.md. Writing my intro...\n\n# Terminal 2 — Antigravity\n> /agent-discussion\nAntigravity: What name should I use? → \"antigravity\"\nAntigravity: File path? → \"./discussion.md\"\nAntigravity: Found existing discussion about \"REST vs GraphQL\". Joining...\n```\n\nOr provide everything inline:\n```\n/agent-discussion agent:claude topic:\"REST vs GraphQL\" file:./discussion.md\n```\n\n## Watch It Happen\n\n```bash\nwatch -n 3 cat discussion.md\n```\n\nThe discussion file grows incrementally. Each agent appends its intro and rounds — never overwrites.\n\n## Discussion File Format\n\n```markdown\n# Discussion: REST vs GraphQL for our new API\n<!-- status:active -->\nStarted: 2026-03-02 10:00\n\n[TURN: claude]\n\n**claude (Intro):** REST and GraphQL address different problems. REST leverages HTTP\nsemantics — caching, status codes, standard tooling — and is broadly understood. GraphQL\nsolves over/under-fetching and consolidates divergent client data needs into one endpoint.\nBoth are production-proven. The decision hinges on consumer diversity and team experience.\n\n[TURN: antigravity]\n\n**antigravity (Intro):** The REST vs GraphQL debate is partly false: REST excels at\nresource-oriented APIs with stable shapes; GraphQL excels when clients have heterogeneous\ndata needs. Key tradeoffs: REST has simpler caching; GraphQL has stronger typing and\nintrospection. Migration cost from REST to GraphQL is low; a poorly designed GraphQL schema\nis expensive to fix.\n\n[TURN: claude]\n\n**claude (Round 1):** Given antigravity's intro agrees the migration cost is low, starting\nwith REST is the pragmatic call. Ship fast, validate consumer needs, add GraphQL when\nthe pain is real — not speculative.\n\n[TURN: antigravity]\n\n...\n\n[AGREED: Start with REST; evaluate GraphQL adoption at 6 months based on real consumer feedback.]\n```\n\n## Installation\n\n### As a Claude Code Plugin\n\nInstall via the Claude plugin system:\n```\n/plugin install github:cyrusxyl/agent-discussion\n```\n\n### Manual Installation\n\nCopy `skills/agent-discussion/SKILL.md` to your project's `.claude/skills/agent-discussion/SKILL.md`.\n\n## Repo Structure\n\n```\nagent-discussion/\n├── .claude-plugin/\n│   └── plugin.json          # Plugin metadata\n├── skills/\n│   └── agent-discussion/\n│       └── SKILL.md         # The skill — protocol, loop, convergence rules\n├── LICENSE\n└── README.md\n```\n\n## Works With Any Agent\n\nAny AI agent that can read and append to files can participate. Tested with Claude Code; compatible with Antigravity, Copilot, Gemini CLI, and any other agent that can follow the protocol in `SKILL.md`.\n\n**Resumable.** Stop an agent, restart it, point it at the same file — it reads the history and picks up exactly where things left off.\n\n## Tips\n\n**One file is all you share.** Both agents need access to the same file path. Use a shared drive, a repo, or just run both agents in the same directory.\n\n**The intro phase matters.** Because each agent writes an unbiased overview before taking a position, debates tend to converge faster — both sides acknowledge the same tradeoffs from the start.\n\n**Deadlock is finite.** If agents can't agree after 12 rounds, the skill declares deadlock and lists the unresolved points. No infinite loops.\n",
  "bytes": 4277,
  "sha": "40fc4705ad47744fcd8a4f17cc40ec934ac2c8ed8fe4b5205f95fa584705d0f2",
  "repo_slug": "cyrusxyl/agent-discussion",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_cyrusxyl_agent_discussion_agent_discussi_039ce799/readme"
}