safe-agentic-workflow  by bybren-llc

AI agent harness orchestrates multi-agent teams using SAFe methodology

Created 9 months ago
320 stars

Top 84.3% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

SAFe Agentic Workflow (SAW) is a production-tested AI agent harness that adapts the Scaled Agile Framework (SAFe) for multi-agent team coordination. It targets teams across various domains (software, marketing, research, etc.) seeking structured, repeatable processes, enhancing autonomy and evidence-based delivery.

How It Works

SAW utilizes a three-layer architecture (Hooks → Commands → Skills) and SAFe principles for AI agent roles and quality gates. It supports multiple AI providers (Claude Code, Gemini CLI, Codex CLI, Cursor IDE) for coordinated orchestration. The "Dark Factory" enables persistent autonomous agent teams via tmux.

Quick Start & Requirements

Setup involves copying provider-specific harness configurations (e.g., .claude/) and customizing placeholders. Prerequisites include a supported AI provider, Git, and necessary API keys/CLIs (e.g., npm install -g @google/gemini-cli). Commands like /start-work initiate tasks. Detailed setup and adoption guides are available at docs/guides/GETTING-STARTED.md, docs/guides/WORKSPACE-ADOPTION-GUIDE.md, and docs/HARNESS_SYNC_GUIDE.md. Gemini CLI documentation is at geminicli.com.

Highlighted Details

  • Features 18 Skills, 24 Commands, and 11 SAFe Agent Profiles.
  • Incorporates a three-layer architecture and SAFe quality gates.
  • Developed over 5 months (2,193 commits, 169 issues), implementing Anthropic research patterns.
  • Supports persistent autonomous teams via "Dark Factory."

Maintenance & Community

Active development is indicated by 2,193 commits and 169 issues over 5 months. Maintained by J. Scott Graham (ByBren, LLC). No specific community channels are detailed.

Licensing & Compatibility

Released under the MIT License, but requires attribution in derivative works per the NOTICE file, which may affect commercial or closed-source use.

Limitations & Caveats

Provider integration maturity varies (Claude Code is deepest). The mandatory attribution requirement is a key consideration. Non-SWE domain adaptations are documented but not production-validated.

Health Check
Last Commit

3 months ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
257 stars in the last 30 days

Explore Similar Projects

Starred by Andrew Ng Andrew Ng(Founder of DeepLearning.AI; Cofounder of Coursera; Professor at Stanford), Jack Lukic Jack Lukic(Author of Semantic UI), and
5 more.

ag2 by ag2ai

0.4%
5k
AgentOS for building AI agents and facilitating multi-agent cooperation
Created 1 year ago
Updated 1 day ago
Feedback? Help us improve.