Skip to main content
This guide walks you through the fastest path to a working Doppels run. You will initialize a Space, use the example Capabilities bundled in the repository, execute a run, and inspect its output — all without needing an AI agent. By the end you will understand the full local workflow: validate, run, inspect.

​
Prerequisites

  • The doppels CLI installed and on your PATH. See the Installation guide if you haven’t done this yet.
  • A local clone of the Doppels repository (only needed for the quickstart example Space):

​
Quickstart walkthrough

1

Initialize a Space

A Space is a project directory that contains your Capabilities, Recipes, and run history. Initialize one in any project folder with:
This creates three directories:For this quickstart, you will use the example Space already present in the cloned repository rather than starting from an empty project.
2

Run the quickstart example

Navigate to the example Space and validate all manifests:
A passing validation looks like:
Now run the greet Capability, passing name as an input and confirming automatically with --yes:
Expected output:
List all runs recorded in this Space:
3

Inspect the run

Copy the run ID from doppels runs list and use it to examine the full run record and its logs.Show the structured run summary:
Stream the step-level logs:
Every step’s stdout and stderr is captured and stored under .doppels/runs/<run-id>/. Nothing is sent off your machine.
4

List your Capabilities

See everything available in the current Space:
Inspect the full contract of a specific Capability:
This prints the Capability’s inputs, outputs, and any linked Recipe — useful before running or sharing.
Add --json to most commands for machine-readable output. For example, doppels runs list --json returns a JSON array of run objects, making it easy to pipe results into other tools or scripts.

​
What’s in the example Space

The examples/quickstart Space contains three Capabilities that exercise different parts of the system:
  • greet — a simple single-step shell Recipe that produces a message string and a receipt.txt artifact. Good for verifying your install.
  • release-pipeline — a four-step Recipe (roughly 1–3 seconds per step) that demonstrates multi-step sequencing and inter-step references. Run it with doppels run capability/release-pipeline --input version=1.2.3 --yes.
  • manual-review — a Capability with no Recipe, fulfilled by passing --output and --evidence flags directly to doppels run. Shows that Capabilities can be satisfied manually without any shell automation.

​
Next steps

Freeze with your agent

Learn how to install the freeze skill and capture a real agent session as a reusable Capability.

Capabilities in depth

Understand the full Capability schema: inputs, outputs, versioning, and manual fulfillment.