Welcome to agentcmd
Build multi-step AI workflows with agentcmd that orchestrate Claude, OpenAI Codex, and Google Gemini to automate complex development tasks.
Think of it as GitHub Actions for AI agents. You define workflows with phases and steps, then agentcmd handles execution, monitoring, and artifact management.
What is agentcmd?
agentcmd is a platform for automating development workflows with AI agents. It consists of two components:
- Web Application - Visual workflow management, real-time monitoring, and run history
- Workflow SDK (
agentcmd-workflows) - TypeScript library for defining workflows
Why agentcmd?
How It Works
What happens in each step:
- Write Spec - Generate feature specification using
/cmd:generate-feature-spec - Run Workflow - Trigger automated workflow execution via UI or CLI
- Setup (system) - Create git branch/worktree, prepare environment
- Execute Workflow - A user defined workflow is executed, an example would be:
- Implement -
step.agent()calls Claude/Codex with/cmd:implement-specto write code - Review -
step.agent()calls Claude with/cmd:review-spec-implementationto validate changes - Test - Run tests, type-checks, linting via
step.cli()
- Implement -
- Finalize (system) - Create PR with
step.git(), merge, and deploy
Quick Example
Here's a complete workflow matching the diagram above:
import {
buildSlashCommand,
defineWorkflow,
type CmdImplementSpecResponse,
type CmdReviewSpecImplementationResponse,
} from "agentcmd-workflows";
export default defineWorkflow(
{
id: "implement-review-workflow",
name: "Implement Review Workflow",
description: "Implements a spec file and reviews the implementation",
phases: [
{ id: "implement", label: "Implement" },
{ id: "review", label: "Review" },
{ id: "test", label: "Test" },
],
},
async ({ event, step }) => {
const { workingDir, specFile } = event.data;
await step.phase("implement", async () => {
const response = await step.agent<CmdImplementSpecResponse>(
"implement-spec",
{
agent: "claude",
json: true,
prompt: buildSlashCommand("/cmd:implement-spec", {
specIdOrNameOrPath: specFile,
format: "json",
}),
workingDir,
}
);
return response;
});
await step.phase("review", async () => {
const response = await step.agent<CmdReviewSpecImplementationResponse>(
"review-spec-implementation",
{
agent: "claude",
json: true,
prompt: buildSlashCommand("/cmd:review-spec-implementation", {
specIdOrNameOrPath: specFile,
format: "json",
}),
workingDir,
}
);
return response;
});
await step.phase("test", async () => {
await step.cli("test", {
command: "pnpm test && pnpm check-types",
workingDir,
});
});
}
);Core Concepts
How It Works
Understand the agentcmd architecture and workflow execution
Workflow Definitions
Multi-step processes orchestrated by Inngest
Agents
AI CLI tools (Claude/Codex/Gemini) that execute your workflows
Slash Commands
Reusable type-safe commands that template your engineering
Spec-Driven Development
Feature specifications that drive implementation workflows
Sessions
Agent conversation history enabling resumable workflows
Get Started in 5 Minutes
Install agentcmd
npx agentcmd installStart the server
npx agentcmd install
npx agentcmd startOpen the UI
open http://localhost:4100
open http://localhost:8288Create your first project
Add your codebase, run an example workflow, and see AI agents in action.
What You'll Build
Recursive Workflows
Advanced workflows that implement and review recursively
Type-Safe Slash Commands
Reusable prompts with type-safe args
Context Sharing
Share data between workflow steps
Ready to Start?
Jump into the Getting Started guide to install agentcmd and build your first workflow in under 30 minutes.
Or explore:
- Core Concepts - Understand workflows, phases, and steps
- Step Types Reference - Complete API for all 9 step types
- Examples - Real-world workflow templates
Need Help? Open an issue on GitHub.