Workflow atlas

Software factory

Delegate typed issues to resident agents for CI-gated implementation, adversarial review, and automatic merge.

Rendering workflow…

Workflow step

Configure

Open the dashboard to install the GitHub App and manage the workflow bundle in your account or organization .agents repository. Managed resident execution is planned.

Install the agents

This workflow uses engineer. Install its agent bundle, preview the repository changes, then open a pull request or commit them to the default branch.

Run it

This delivery workflow begins with an issue created by a human or agent. Coding work is assigned to a resident agent running on infrastructure you operate. Managed resident deployment is planned, not available today.

How the declaration maps to what actually runs:

Declared step What runs Identity
prepare_issue, assign Accept a human- or agent-created issue, categorize it only when it is untyped, and assign or confirm coding work to a resident. human or agent → resident
worktree, research, implement, draft The resident creates a worktree, researches, implements, iterates, and pushes a draft pull request. resident identity or ai-outfitter[bot]
ci, ready, auto_merge Required CI gates readiness. After it passes, the resident undrafts the pull request and enables GitHub auto-merge. resident identity or ai-outfitter[bot]
review, revise CODEOWNERS requests an adversarial-review agent with a GitHub identity distinct from the implementer. Requested changes return to implementation, CI, and review until approval. distinct review agent
merge GitHub auto-merges only after required CI and the required agent approval. GitHub

Run

  1. Open an issue with acceptance criteria a reviewer can check.
  2. Assign a resident agent running through your current harness or infrastructure.
  3. A draft pull request appears and stays draft until required CI passes.
  4. CODEOWNERS requests the distinct adversarial reviewer. Requested changes repeat revision, CI, and review; approval lets GitHub auto-merge.

Repository controls drive the process

Repository control How it enables the agentic process
Issue templates Give agents a clear starting point: a typed request, the context they need, and acceptance criteria they can verify.
CODEOWNERS Requests the right agent or human reviewers when a pull request becomes ready, starting adversarial review without another handoff.
Branch protection and auto-merge Hold the branch until required reviews and checks pass, then merge it without another manual step.

See it on a real repository

The reference run lives in ai-outfitter/factory-demo-target, a deliberately tiny project with no agent configuration: