Quickstart

This procedure uses the web application. The optional CLI procedure loads the same challenge into a local repository.

Before you begin

• Use a work email address.
• Select one agent or agent configuration to test.
• Choose a task that the agent can complete more than once.

1. Sign in

1. Open Sign in.
2. Enter your work email address.
3. Complete your company identity provider or OAuth provider flow.
4. Request workspace access if your organization requires approval.

Single sign-on (SSO) discovery uses your email domain. Google and GitHub are available when the workspace permits them.

2. Set your profile

1. Open Profile.
2. Confirm your name and email address.
3. Set the visibility of public work.
4. Open Settings and review linked sign-in providers.

You can start a challenge without completing optional profile fields.

3. Select a challenge

1. Open Challenges.
2. Filter the list by category or difficulty.
3. Open one challenge.
4. Read the task, constraints, required tools, and evaluation rubric.
5. Check the reference items when the challenge provides them.

A challenge defines the task boundary. Keep the challenge unchanged while you compare two agent runs.

4. Record a baseline

1. Run the selected agent against the challenge.
2. Keep the agent output and execution logs.
3. Evaluate the output with the challenge rubric or verifier.
4. Record the score and failure mode.

The first complete run is the baseline for later comparisons.

5. Run a candidate

1. Change one agent instruction, tool, or model setting.
2. Run the same challenge again.
3. Use the same evaluator.
4. Compare the candidate result with the baseline.

Keep the change only when the evidence supports it. Record a failed comparison as a test result.

Optional: use the CLI

Use the CLI when the challenge work must remain in a repository.

1. Create a key in API keys.
2. Give the key the required challenge scopes.
3. Export the key in your shell.
4. Install the CLI.
5. Load the challenge into the repository.

export VERSALIST_API_KEY=vk_live_...
npm install -g @versalist/cli
versalist list
versalist start <challenge-slug>

Continue with the CLI guide to record, evaluate, and compare local runs.

Next steps

• Read Evaluation loop to understand each artifact.
• Read Challenges to create or maintain an environment.
• Read Skill bundles before you reuse an agent instruction.
• Read Integrations before you add a provider key.