Skip to main content

Agent quickstart

Use this flow when an agent needs to evaluate an existing CLI.

Goal

1. Install and initialize

inter init creates the Intermesh workspace and installs the managed planning Skill.

2. Select the CLI

From a repository that contains one CLI package:
If there is more than one package, Intermesh returns structured choices. The agent reruns setup with the requested explicit package directory, package manager, and build script. For an npm package, tarball, or remote Git repository, use the source-specific setup options in Bring an existing CLI.

3. Prepare the candidate

Use the installed inter-eval-planning Skill with the coding agent. Review .intermesh/evaluations/default/eval.json: source, tasks, resource IDs, expected outcomes, target account, allowed hosts, and scenario selection. Prepared artifacts and detailed execution evidence remain in private machine-local storage.
Preparation freezes and inspects the artifact. If the plan is missing or invalid, use the installed planning Skill to create or repair it, then run:

4. Run after authorization

Only run a reviewed plan with the dedicated, restricted credential it names.

5. Return evidence

Read the report and return the evidence, not only the completion status.