Skip to content

CLAUDE.md

Latest commit

 

History

History
75 lines (55 loc) · 2.2 KB

CLAUDE.md

File metadata and controls

75 lines (55 loc) · 2.2 KB
Raw

AgentLoop

Platform-agnostic AI agent: monitors chat platforms, responds using Claude Code subprocess.

Tech Stack

  • TypeScript (strict, ESM), Node.js 22+
  • Claude Code CLI (npx @anthropic-ai/claude-code) as subprocess
  • dotenv for environment variables
  • Build: tsc → JS. Dev: tsx

Commands

npm run dev        # Run with tsx (dev)
npm run build      # Compile TypeScript
npm start          # Run compiled JS
npm run test:e2e   # E2E test (see below)

Running E2E Tests

E2E tests require a running agentloop instance connected to Slack:

# Terminal 1: Start the agent
npm run dev

# Terminal 2: Run tests
npm run test:e2e

The tests post messages to the configured Slack channel and verify the agent responds with ✅ reactions.

Project Structure

src/
  cli.ts                   # CLI entry point (agentloop serve)
  agent.ts                 # Wraps ClaudeCodeExecutor with system prompt
  claude-code-executor.ts  # Spawns Claude Code subprocess
  config.ts                # Config loader with env var substitution
  platforms/
    slack.ts               # Slack adapter: polling, reactions, thread replies

Code Style

  • ESM imports (no CommonJS)
  • Strict TypeScript, explicit types on public APIs
  • Minimal dependencies — prefer stdlib
  • Error handling: catch per-message, never crash the loop
  • Config: JSON with ${ENV_VAR} substitution

Architecture Notes

  • Claude Code runs as subprocess with cwd set to workspace directory
  • Workspace cwd enables: CLAUDE.md loading, filesystem access, git awareness
  • MCP servers passed via --mcp-config flag to Claude Code
  • State via Slack reactions (✅) — no database needed
  • Platform-agnostic design: add adapters in src/platforms/

Infrastructure

  • Docker multi-stage build
  • Use tofu not terraform for IaC

Release Process

  1. Bump version in package.json
  2. Commit: git commit -m "chore: bump version to X.Y.Z"
  3. Tag: git tag -a vX.Y.Z -m "Release vX.Y.Z"
  4. Push: git push origin main && git push origin vX.Y.Z
  5. Create GitHub release: gh release create vX.Y.Z --title "vX.Y.Z" --notes "..."
    • CI publishes npm package (OIDC provenance) and Docker image to ghcr.io/fullstackhouse/agentloop