Level CLI can generate an Agent Skill that teaches a compatible coding agent how to use your installed version of lvl. You choose the coding agent, model, and token provider. Level supplies the CLI commands and the generated SKILL.md reference.

You need:

Run lvl --help, then run lvl auth status to confirm that the CLI is available and signed in to the intended Level organization.

Use these commands to inspect what the coding agent will receive:

The skill is generated from the command tree in your installed CLI, so its commands and output definitions match that version of lvl.

Run lvl skill install.

The canonical skill is written to ~/.agents/skills/level/SKILL.md. This location is used by Codex, Cursor, Gemini CLI, and other agents that read the standard ~/.agents/skills directory.

If Claude Code is already configured on the computer, lvl also installs or links the skill at ~/.claude/skills/level. When symbolic links are unavailable, the installer copies the skill instead.

Use a project-level installation when you want the skill to travel with one repository or working directory.

The command writes the canonical skill to ./.agents/skills/level/SKILL.md. It also installs or links the Claude Code location at ./.claude/skills/level.

Review your coding agent's documentation to confirm which Agent Skills directory it reads.

Start a new coding-agent session in the same user or project context after installing the skill. Give the agent a clear target, expected result, and permission boundary.

The following prompts use blockquotes so they wrap on smaller screens. Replace the sample names with resources from your Level account.

Use lvl to list every device in the Servers group. Return the device IDs, names, operating systems, online status, active alerts, and available updates. Do not make changes.

This shows how an agent can combine device, alert, and update data into one fleet summary.

Use lvl to inspect web-01 and determine its operating system. Draft a read-only disk-usage script for that operating system. Show me the complete script, target device ID, shell, run-as setting, timeout, and exact lvl command. Do not create or run the script.

After reviewing the script, you can ask the agent to save it with lvl script create or run it with lvl script run.

Use lvl skill describe automation to review the current automation commands and specification format. Draft a new automation named Disk Space Triage that targets the Servers group, runs a read-only disk diagnostic, and records the output. Write the proposed YAML specification to disk and explain every trigger, action, variable, and target. Do not create the automation.

After reviewing the specification, ask the agent to create it with lvl automation create -f AUTOMATION_SPEC.

Export the Patch Servers automation with lvl automation spec. Propose the smallest YAML change needed to add a wait-for-approval step before patch installation. Show me the diff, then run lvl automation apply with --dry-run. Do not apply the change.

This workflow lets the agent inspect the current full-state specification, edit it, and preview how it would reconcile with the live automation.

Use lvl skill describe monitor policy and lvl skill describe monitor add to inspect the current monitor-policy and monitor payload formats. Draft a Server Health policy for the Servers tag with CPU, memory, and disk-free monitors. Include the proposed thresholds, durations, severities, alert recipients, and target tags. Write the YAML specification to disk and explain every setting. Do not create the policy.

After review, ask the agent to create the policy with lvl monitor policy create -f POLICY_SPEC.

Monitor-policy creation from a specification is also a non-transactional chain. A failure can leave a partially created policy, so review the returned policy ID and correct it with lvl monitor policy apply -f POLICY_SPEC if needed.

Export the Server Health monitor policy. Add a read-only script monitor that checks the application health endpoint. Show me the proposed script and policy diff, then run lvl monitor policy apply with --dry-run. Do not apply the change.

Monitor-policy apply reconciles the live policy with the file. Monitors missing from the specification can be deleted, so the agent should review the full exported policy rather than constructing a partial update.

Inspect the Server Health monitor policy and the Restart Web Service automation. Determine whether the current monitor type supports the intended remediation workflow. Draft the smallest policy and automation changes needed to alert first and run the existing remediation only after approval. Show both diffs and dry-run plans. Do not apply either change.

This prompt makes the agent inspect the installed CLI's current monitor and automation schemas before proposing a connected workflow.

Use lvl to inventory scripts, automations, monitor policies, tags, device groups, and repository files related to patching. Identify duplicate names and ambiguous references. Propose a consistent grouping and naming plan. Do not rename, move, or delete anything.

A read-only inventory is a useful first step before asking an agent to reorganize shared resources.

Check these details before allowing a proposed command to run:

Permanent delete commands require confirmation. In a non-interactive session, commands that require confirmation fail unless --yes is supplied. Some actions, including device restart and shutdown, request the action immediately and do not have a confirmation prompt.

The skill and command reference are generated from your installed CLI version. After upgrading lvl, run lvl skill install again.

For a project-level installation, run lvl skill install --project.

This keeps your coding agent's command index, flags, and JSON output definitions aligned with the installed binary.

Reinstall the skill with the same lvl binary that the coding agent will call. If multiple lvl binaries are on your PATH, run lvl --help, then use lvl skill describe COMMAND to inspect the current command.

The installed CLI and lvl skill describe are the source of truth for the available commands and flags.