Introducing AWF

March 14, 2026 in announcements by Alex4 minutes

For the past few months, I’ve been working on a side project called AWF (AI Workflow Framework). The code for the main CLI tool is now available on GitHub for everyone — including a dedicated Claude skill.

Why I Built This

I wear two hats: tech enthusiast and company owner. Managing multiple Claude sessions with various agents simply doesn’t scale with daily responsibilities. I wanted a way to test my ideas and improve my workflow without sacrificing my standards — I’m a firm believer in TDD, QA, and high engineering standards.

On the AI side, three things kept frustrating me:

Efficiency — I don’t want to waste tokens. If I ask an agent to perform TDD, I want real TDD.
Determinism — I want to call my CLI tools because they are predictable. No crossing fingers hoping an agent calls the right MCP tool.
Flexibility — I want to use Claude, Gemini, or any other model without arbitrary limitations.

That’s why I built AWF in Go. It’s a workflow engine designed to orchestrate CLI tools — including LLMs like Claude, Gemini, and Codex.

What AWF Is (and Isn’t)

AWF isn’t another “magic” AI wrapper. It’s a professional tool for those who know how to manage their context window and want to build deterministic, industrial-grade workflows using CLI outputs.

With AWF, you design your workflow through discrete steps. Each step can run a CLI program, a shell script, or an AI agent. You can:

Pass parameters as options
Capture stdout for the next step
Abort the workflow on stderr
Define transitions, pre/post-events, loops, and retries
Run steps in parallel with configurable strategies (all_succeed, any_succeed, best_effort)

The better you know your CLI basics, the more freedom you have to build complex systems.

The Agent Step

There is a specialized step type for AI interactions. You provide a prompt — which supports variables via Go templates — and AWF executes it according to your design:

analyze:
  type: agent
  provider: claude
  prompt: |
    Review this code for bugs, security issues, and improvements:
    {{.states.read.Output}}
  output_format: json
  options:
    model: sonnet
  timeout: 120
  on_success: report
  on_failure: {message: "Analysis failed: {{.states.analyze.Output}}"}

Want to fail fast? No problem. Continue on error? Easy. Call a nested workflow? That’s actually how I update the AWF skill on every new PR.

Features at a Glance

State machine execution — Define workflows as state machines with conditional transitions based on exit codes, command output, or custom expressions
Multi-agent orchestration — Claude, Gemini, Codex, and any OpenAI-compatible API (Ollama, vLLM, Groq) in the same workflow
Conversation mode — Multi-turn conversations with native session resume, context injection, and token tracking across all turns
External prompt files — Load prompts from .md files with full template interpolation and local override support
External script files — Shebang-based interpreter dispatch with template interpolation
Loop constructs — For-each and while loops with full context access
Retry with backoff — Exponential, linear, or constant backoff strategies
Parallel execution — Run multiple steps concurrently with configurable strategies
Sub-workflows — Invoke workflows from other workflows with input/output mapping
Built-in plugins — GitHub operations, HTTP calls, desktop notifications
Audit trail — Structured JSONL logging with secret masking
Dry-run and interactive modes — Preview or step through execution before committing
XDG support — Workflows can be defined locally or globally with override support

Getting Started

Install AWF:

curl -fsSL https://raw.githubusercontent.com/awf-project/cli/main/scripts/install.sh | sh

Or via Go:

go install github.com/awf-project/cli/cmd/awf@latest

Then create your first workflow:

awf init
awf run example

Read the documentation to learn more, or check out the AWF organization on GitHub for plenty of examples to help you get started.

What’s Next

AWF is already working great for my own projects, but I want to keep improving it. I’d love to get feedback on the code — especially since AWF is now built using its own workflows.

If you’re a Go developer, I’d highly appreciate your insights. Check out the tests, the examples, and don’t hesitate to open an issue.

AWF — AI Workflow Framework CLI

Title here