Skip to main content
devhelm skills install auto-detects installed hosts and copies the bundle to the right location. This page documents each supported host — where files land, how to authenticate, and how to verify. If your host isn’t listed but follows the Anthropic agent-skill convention (a directory of SKILL.md files loaded at session start), use --target generic --path /your/path and point it at the host’s skills directory.

​
Cursor

Project-scoped (recommended) — skills committed with your repo, so every contributor gets them:
This writes to .cursor/skills/. The skill files themselves are safe to commit so every contributor gets them. Global — available in every workspace:
This writes to ~/.cursor/skills/. Authenticate the CLI once so the skill’s shell calls succeed:
Restart Cursor after the first install so it re-scans the skills directory.

​
Claude Code

Claude Code reads skills from ~/.claude/skills/ by default:
To scope skills to a project, pass --path ./.claude/skills:
Authentication is the same as Cursor — Claude Code invokes the devhelm binary and inherits your shell’s environment.

​
Claude Desktop

Claude Desktop imports skills as uploaded bundles (zip files), not directory symlinks. Generate the bundle:
Then in Claude Desktop: Settings → Skills → Add skill → Choose file and select the zip. Do this once per release of the CLI; running devhelm skills bundle again gives you a fresh zip. Because Claude Desktop sandboxes skill execution, set your API token in Claude Desktop’s skill-level environment configuration rather than your shell.

​
Generic / other hosts

The bundle is a plain directory of SKILL.md files plus references/ — any host that understands Anthropic’s skill convention can load it.

​
What gets installed

Each target receives four skill directories:
The _generated/*.fields.md files are regenerated from docs/openapi/monitoring-api.json on every CLI release, so field-level schema is always in sync with the live API.

​
Authentication

The skills themselves don’t store credentials — they shell out to devhelm, which resolves the API token in this order:
  1. --api-token flag (not used by the skills).
  2. DEVHELM_API_TOKEN environment variable.
  3. ~/.devhelm/contexts.json (populated by devhelm auth login).
The starter API key shown on the setup-complete screen (named Default) works fine for all four skills. It’s an ordinary API key — rename, rotate, or delete it like any other. If you want a separate key for agent use, create one in Settings → API Keys; all keys have full access to the organization, so the isolation is about auditability, not capability.

​
Verify

Reports:
  • Whether the bundled skills are present.
  • Whether your CLI is authenticated.
  • Whether skills are installed to at least one agent host directory.
  • Which agent hosts were detected on this machine.
A typical healthy output:

​
Updating

Skills are versioned with the CLI. To upgrade, upgrade the CLI and re-run install:
devhelm skills install is idempotent — it replaces existing files with the new version. If you’ve edited a skill locally, save your changes before re-running.

​
Uninstalling

Delete the skill directories. devhelm doesn’t track install locations separately from what doctor shows, so removing the files is sufficient: