Quickstart
August 27, 2026 ยท View on GitHub
This guide creates a reproducible project-local AI asset setup. It does not require a private registry, a maintainer checkout, or a machine-specific configuration.
Prerequisites
- Python 3.10 or later
- Git
- A supported AI coding runtime, such as Codex, Claude Code, Cursor, Kiro, or dsh
Install
python -m pip install --upgrade harness-ai-kit==0.18.18
harness-ai-kit --version
Initialize Your Local Configuration
harness-ai-kit init
This creates or updates ~/.harness-ai-kit/config.yaml. It is the shared
location for user-specific endpoints and credentials; do not write those
values into a project manifest, lockfile, or Skill.
For a registry-backed public catalog, configure the public registry endpoint in that file. You can also use a public Git source directly, as shown below.
Create A Project And Add A Public Skill
mkdir my-agent-project
cd my-agent-project
harness-ai-kit init-project --runtime codex
harness-ai-kit add skill https://github.com/anthropics/skills/tree/main/skills/skill-creator
init-project writes harness-ai-kit.yml, which declares what the project
wants. add resolves the source, writes harness-ai-kit.lock, and installs
the selected Skill into the configured runtime directory.
Verify The Result
harness-ai-kit sync
harness-ai-kit doctor
For the default Codex project runtime, the installed Skill is under
.agents/skills/. The manifest and lockfile are the files to commit; the
runtime directory is a materialized working copy.
Next Steps
- See the CLI reference for command options.
- See usage scenarios for choosing Skills and Loops.
- See concepts for manifests, locks, sources, and runtime materialization.
- Use troubleshooting when a source, dependency, or runtime check fails.