Agents

Multi-agent support — Claude Code, OpenCode, and GitHub Copilot CLI behind one workflow runtime.

Atomic doesn’t replace your coding agent. At each stage it spawns the agent’s own CLI and talks to it via that agent’s SDK — so file editing, tool use, MCP setup, hooks, and context handling all keep working as they do today.

Supported agents

Agent	Chat	Workflow flag
Claude Code	atomic chat -a claude	-a claude
OpenCode	atomic chat -a opencode	-a opencode
GitHub Copilot CLI	atomic chat -a copilot	-a copilot

Each agent gets its own configuration directory (.claude/, .opencode/, .github/), skills, and context files — all managed by Atomic. See agent-specific files.

Switching agents

Use the same workflow across agents by pinning a variant in .for(...):

// src/workflows/review/claude.ts
export default defineWorkflow({ name: "review", description: "..." })
  .for("claude")
  .run(async (ctx) => { /* ... */ })
  .compile();

// src/workflows/review/copilot.ts
export default defineWorkflow({ name: "review", description: "..." })
  .for("copilot")
  .run(async (ctx) => { /* ... */ })
  .compile();

Per-agent session APIs are slightly different:

Agent	How to send a prompt
Claude	await s.session.query(prompt)
Copilot	await s.session.send({ prompt })
OpenCode	await s.client.session.prompt({ sessionID: s.session.id, parts: [{ type: "text", text: prompt }] })

See stages for the full callback surface.

When to use -a on your own CLI

If your worker file imports a workflow pinned to one agent (import workflow from "./claude/index.ts"), there’s nothing to disambiguate — no -a flag. Only reach for -a/--agent when one CLI dispatches across workflows that exist in multiple agent variants. The atomic CLI itself uses -a for that reason: its built-in registry has cross-agent variants of ralph, deep-research-codebase, and open-claude-design.

Agent reference

Claude Code

Claude Code CLI reference.

OpenCode

OpenCode server reference.

GitHub Copilot CLI

Copilot CLI usage reference.

Permissions

How Atomic bypasses prompts in workflows — and why that means containerization.