Caution
Agency was a research project. If your name is not Michael Uloth, do not use it.
No support or warranty is provided. Use at your own risk.
Agency is a meta-harness that runs other AI coding agents through automated multi-step workflows.
- analyze — scan a codebase for issues and file them in the tracker
- fix — pick up tracked issues (or ad-hoc specs), run implement→review loops, open PRs
- watch — continuous loop: fix ready issues, analyze when the queue is low, sleep
- plan — interactive session to explore a codebase and produce a structured plan
- ralph — iterative fresh-eyes refinement toward a goal, committing per iteration
graph TD
CLI["<b>Entrypoints</b><br/>CLI"]
CLI --> Pipelines["<b>Pipelines</b><br/>analyze · fix · plan · ralph · watch"]
Pipelines --> Strategies["<b>Loop Strategies</b><br/>AntagonisticStrategy · RalphStrategy"]
Pipelines --> Ports["<b>Ports</b><br/>AgentBackend · VCSBackend · IssueTracker"]
Ports --> Adapters["<b>Adapters</b><br/>ClaudeCliBackend · GitBackend · GitHubTracker"]
See ARCHITECTURE.md for more detail.
Optionally add a .agency/config.yml in your project root to customize behavior.
All fields are optional — sensible defaults are built in.
See the Config type in ARCHITECTURE.md for the full schema.
- Case Statement: Building a Harness + workos/case by Nick Nisi
- Harness engineering: leveraging Codex in an agent-first world by Ryan Lopopolo
- Effective harnesses for long-running agents by Justin Young
- Harness design for long-running application development by Prithvi Rajasekaran
- Harness Engineering by Birgitta Böckler
- Relocating Rigor by Chad Fowler
- Skill Issue: Harness Engineering for Coding Agents by Kyle Mistele
- Engineer the Harness by Mitchell Hashimoto