This guide is the recommended first-time path for getting Agent Workflow Framework (AWF) installed and working against your first target repository.
By the end of this guide you will have:
- installed AWF once on your machine
- bootstrapped a target repository with the
.wi/artifacts - enabled the adapters already available on your machine
- verified the repo setup with AWF diagnostics
- started your first story
Make sure the machine has:
- Node.js and npm
- Git
- PowerShell on Windows or Bash on macOS/Linux
- at least one supported adapter CLI if you want AWF to launch agents directly:
- Codex
- Claude
- Gemini
- GitHub Copilot CLI exposed as
copilot
For dependency details, see DEPENDENCIES.md.
From the framework repo root, run:
./installers/install-framework.ps1Or on Bash:
./installers/install-framework.shThis installs the framework into a shared location so you can reuse it across many repos.
On Windows, install-framework.ps1 also adds the AWF bin directory to your user PATH when needed. Open a new terminal after install if awf is not available in the current session yet.
On Bash, the installer does not edit your shell profile for you, so add ~/.awf/bin to PATH if you want to launch awf from any terminal.
Move to the repository where you want AWF to create and manage the .wi/ working artifacts.
If you want to inspect what tooling AWF can already detect before changing anything, run:
./bin/awf.ps1 tooling-status --repo C:\repo./bin/awf tooling-status --repo /repoUse the guided bootstrap flow for your first repo:
./bin/awf.ps1 bootstrap-repo --repo C:\repo --enable-installed./bin/awf bootstrap-repo --repo /repo --enable-installedThis is AWF's closest equivalent to an installation wizard. In one pass it:
- initializes
.wi/ - detects the supported CLIs already installed on the machine
- updates
.wi/config.json - enables the installed adapters you already have
- runs a validation pass so you can catch setup problems early
If you prefer the installer wrapper instead of calling the CLI directly, use:
./installers/bootstrap-repo.ps1 -RepoPath C:\repo -EnableInstalled./installers/bootstrap-repo.sh --repo /repo --enable-installedAfter bootstrap, check these files in the target repo:
.wi/config.json.wi/story.json.wi/tasks.json.wi/acceptance.json.wi/runtime/orientation.json
Not every file will contain active work yet, but the structure should now exist and be ready for the workflow.
Run:
./bin/awf.ps1 tooling-status --repo C:\repo./bin/awf tooling-status --repo /repoIf AWF reports missing tools, install the missing CLI and rerun:
./bin/awf.ps1 configure-tooling --repo C:\repo --enable-installed./bin/awf configure-tooling --repo /repo --enable-installedAfter bootstrap, the guided shell is the recommended entry point for all day-to-day work:
./bin/awf.ps1 shell --repo C:\repoThe shell shows the current workflow state, recommends the next safe action, and walks you through intake, planning, implementation, review, and verification without needing to remember individual flags.
The shell will prompt you through story intake. If you prefer to run it directly from the command line, here are the equivalent commands.
From Markdown:
./bin/awf.ps1 start-story --repo C:\repo --story-file C:\repo\story.mdFrom Jira:
./bin/awf.ps1 start-story --repo C:\repo --jira-url https://your-site.atlassian.net/browse/ABC-123If the story depends on a design doc, include it so the AI intake pass can compare story vs design:
./bin/awf.ps1 start-story --repo C:\repo --story-file C:\repo\story.md --design-file C:\repo\design.md --aiIf intake reports needs_clarification, use the interactive loop:
./bin/awf.ps1 clarify-story --repo C:\repo --interactiveOnce intake is stable, the shell will guide you through the rest. The equivalent direct commands are:
./bin/awf.ps1 plan --repo C:\repo
./bin/awf.ps1 run-next --repo C:\repo
./bin/awf.ps1 review --repo C:\repo
./bin/awf.ps1 verify-story --repo C:\repo- Start with
bootstrap-repoinstead ofinit-repounless you deliberately want the bare minimum scaffold. - Use
--enable-installedon the first pass so AWF only enables tooling that actually exists on your machine. - Run
tooling-statusbefore and after setup when you want a quick diagnostic without guessing. - Keep
auto-commitoff until you are comfortable with the workflow in your repo.
Use these references next: