Install where you need it. Check what is loaded.

Prepare the host and runtime.

Choose an agent tool that supports loading skills as your host. The commands below require Node.js and npm versions supported by the skills CLI you use.

Skills have additional prerequisites:

Match installation scope to how you work.

One scope is usually enough. If several copies have the same name, confirm which location is actually loaded.

Install only the packages you need now.

Run the required commands in a terminal at the target project's root:

npx skills add yuzio-ai/skills --skill cohestra
npx skills add yuzio-ai/skills --skill cadra
npx skills add yuzio-ai/skills --skill vistra
npx skills add yuzio-ai/skills --skill relatra

Each line installs one skill. Install Cohestra as well when using Cadra for engineering execution.

For a user-level Cohestra installation:

npx skills add yuzio-ai/skills --skill cohestra -g

Use -a to select a host explicitly. For example, to install for Codex:

npx skills add yuzio-ai/skills --skill cohestra -a codex

Use the actual paths and results returned by the installer as your basis.

Check that the host can find it.

List project-level installations:

npx skills list

List user-level installations:

npx skills list -g

Return to the host, confirm that the entry is visible, and check release.json in the directory it actually loads.

Skill installation paths and project execution directories serve different purposes. Installation does not create business tasks automatically. The project's .yuzio/ directory stores later execution materials.

Specify the skill and scope to update.

To update project-level Cohestra, run this at the corresponding project root:

npx skills update cohestra -p

To update user-level Cohestra:

npx skills update cohestra -g

Replace the name for other skills. After updating, check the actual loaded path, version, and installation record again.

Use the current installer's help for supported options:

npx skills --help

If your installer cannot update by skill name and scope, address its version first while preserving the intended update scope.

Keep the complete package and the existing records.

For manual installation, place the complete skill directory in a location supported by the host, including references, templates, scripts, and hidden files. Later replacements must also explicitly replace the complete package.

For copies installed from local paths, an update by name still needs verification that the content changed. Reinstall from the original source and scope if needed. Continue using the installer for copies it manages.

After a skill update, preserve the original task's valid decisions, acceptance criteria, and historical handoffs.

If the project protocol is unsupported, preserve the working state and follow the skill's handling rules. Updating a skill does not by itself migrate project records.

Cadra protocol 0.8 reads historical 0.6/0.7 Tasks and Runs without writing them; continue unfinished old Runs with a matching older version. Only when the project has no unfinished Run and the user explicitly authorizes upgrading a specified Task may add back up and upgrade it, followed by a new ready check. Preserve old Runs and handoffs. Explicitly choose implementation for old small-change, or a legal new flow for old full / feature. A protocol upgrade does not reopen completed tasks; tasks selected for execution must pass admission again.

Confirm the entry. Make your first invocation.

Once installation is complete, start with a well-scoped task.