@reaatech/agent-runbook-agent

AI agent layer for the Agent Runbook Generator. Provides LLM provider abstraction for Anthropic Claude, OpenAI, and Google Gemini, plus configurable prompt templates for automated code analysis and runbook generation.

Installation

npm install @reaatech/agent-runbook-agent
# or
pnpm add @reaatech/agent-runbook-agent

Feature Overview

• Multi-provider support — Anthropic Claude, OpenAI (GPT), and Google Gemini with unified interface
• Provider adapter — automatic model formatting and response parsing per provider
• Configurable prompts — pre-built templates for 8 analysis and generation tasks
• Custom prompts — programmatic creation of custom prompt templates
• Cost and performance — supports temperature, max_tokens, and rate limit configuration
• Dynamic SDK loading — provider SDKs loaded on demand to minimize bundle size

Quick Start

import { createAnalysisAgent, generatePrompt } from "@reaatech/agent-runbook-agent";

const agent = createAnalysisAgent({
  provider: "claude",
  model: "claude-opus-4-5",
  apiKey: process.env.ANTHROPIC_API_KEY,
  temperature: 0.3,
});

const insights = await agent.analyzeRepository(analysisContext);
const failureModes = await agent.identifyFailureModes(analysisContext);
const section = await agent.generateRunbookSection("alerts", analysisContext);

API Reference

Analysis Agent

| Export | Kind | Description | |--------|------|-------------| | AnalysisAgent | class | LLM-powered repository analysis. Creates an agent for analyzing code patterns, identifying failure modes, and generating runbook sections. | | createAnalysisAgent | function | (config?: Partial<AgentConfig>) => AnalysisAgent — factory with sensible defaults |

AgentConfig: { provider: string; model?: string; apiKey?: string; baseUrl?: string; temperature?: number; maxTokens?: number }

Instance Methods

| Method | Returns | Description | |--------|---------|-------------| | analyzeRepository(context) | Promise<AnalysisInsight[]> | Analyze repository patterns for insights | | identifyFailureModes(context) | Promise<string[]> | Identify potential failure modes | | generateRunbookSection(sectionType, context) | Promise<string> | Generate a runbook section |

Provider Adapter

| Export | Kind | Description | |--------|------|-------------| | ProviderAdapter | class | Handles prompt formatting and response parsing for each LLM provider | | createProviderAdapter | function | (config: AgentConfig) => ProviderAdapter — factory |

Instance Methods

| Method | Description | |--------|-------------| | formatMessages(system, user) | Format messages for the target provider | | formatForClaude(system, user) | Anthropic-specific message formatting | | formatForOpenAI(system, user) | OpenAI-specific message formatting | | formatForGemini(system, user) | Gemini-specific message formatting | | parseResponse(provider, raw) | Parse raw provider response into AgentResponse | | getFallbackProvider() | Get fallback provider if primary fails | | supportsStreaming() | Check if provider supports streaming | | getRateLimits() | Get provider rate limits |

Prompt Templates

| Function | Signature | |----------|-----------| | generatePrompt | (type: PromptType, variables: Partial<PromptVariables>) => string | | getPromptTemplate | (type: PromptType) => PromptTemplate | | getSystemPrompt | (type: PromptType) => string | | createPromptTemplate | (type: PromptType, systemPrompt: string, userPrompt: string) => PromptTemplate |

Prompt types: repository-analysis, failure-mode-identification, runbook-alerts, runbook-dashboards, runbook-failure-modes, runbook-rollback, runbook-incident-response, runbook-health-checks.

Related Packages

• @reaatech/agent-runbook — Core types and utilities
• @reaatech/agent-runbook-observability — Logging, tracing, metrics
• @reaatech/agent-runbook-runbook — Runbook assembly pipeline

License

MIT