.codex-plugin/plugin.json

{
  "author": {
    "name": "Krishna Sathvik"
  },
  "description": "Design reliable LLM, tool, MCP, and agent workflows.",
  "interface": {
    "capabilities": [
      "Choose the simplest architecture from deterministic code through multi-agent systems",
      "Design LLM applications, tool-using workflows, autonomous agents, and multi-agent systems",
      "Define safe tool schemas, permissions, retries, idempotency, errors, and approval requirements",
      "Design manager, handoff, router, parallel-worker, evaluator, and orchestrator patterns",
      "Plan MCP tools, resources, prompts, authentication, long-running tasks, and observability",
      "Design conversation state, workflow state, durable preferences, and application memory",
      "Add guardrails, human approvals, least-privilege controls, and prompt-injection defenses",
      "Create agent evaluations for task success, tool use, trajectories, side effects, latency, and cost",
      "Design tracing and observability for model calls, tools, handoffs, approvals, retries, and outcomes",
      "Use current official docs for model APIs, Agents SDK, MCP, tools, guardrails, and platform behavior"
    ],
    "category": "Developer Tools",
    "composerIcon": "./.codex-plugin/assets/composer-icon.svg",
    "defaultPrompt": [
      "Design the simplest reliable LLM or agent architecture for this use case.",
      "Design safe tools, permissions, state, and approvals for this agent.",
      "Review this agent workflow for unnecessary complexity, security risks, and eval gaps."
    ],
    "developerName": "Krishna Sathvik",
    "displayName": "LLM & Agent Builder Copilot",
    "logo": "./.codex-plugin/assets/logo.svg",
    "longDescription": "LLM & Agent Builder Copilot helps you design and implement reliable LLM applications, tool-using workflows, MCP integrations, and agentic systems without adding unnecessary autonomy. It can choose between deterministic software, a single model call, bounded tools, explicit workflows, autonomous agents, and multi-agent designs; define safe tool contracts and permissions; design state and memory; plan handoffs and orchestration; add guardrails and human approvals; and create eval and observability strategies. It favors structured outputs, least privilege, bounded execution, recoverable state, and simple architectures.",
    "shortDescription": "Design reliable AI agents"
  },
  "keywords": [
    "llm",
    "agents",
    "mcp",
    "tools",
    "guardrails",
    "evals",
    "orchestration"
  ],
  "name": "llm-agent-builder-copilot",
  "skills": "./skills",
  "version": "0.1.0"
}