Agent skills
Anchor ships three agent skills: short instructions an AI coding agent
loads when a task needs them. With them installed, an agent in a project that uses Anchor runs
your app with anchor run instead of writing a .env, never puts a secret value or a password
into a command, migrates config files with a dry run first, and gives CI a narrow token.
Install
Section titled “Install”In the project where your agent works:
npx skills add sotarak/anchor --skill anchor-cli --skill anchor-migrate --skill anchor-ciThe skills CLI asks which agents to install for (Claude
Code, Codex, Cursor and others) and whether to install for this project or for your user. Add
-g for every project, -a claude-code to pick the agent, or -y to accept the defaults.
Name the three skills: the Anchor repository also holds the skills its own contributors use, and
npx skills add sotarak/anchor --list shows those too.
What each skill does
Section titled “What each skill does”| Skill | The agent uses it when | It makes the agent |
|---|---|---|
anchor-cli |
A project has .anchor.json, an app is missing a variable, or you ask about secrets |
Run apps with anchor run, read keys and metadata without --reveal, hand you anchor secrets set KEY to type a value into a hidden prompt, ask you to run anchor login instead of using a password, export only to gitignored files |
anchor-migrate |
Moving .env, YAML or JSON config into Anchor |
Map services to folders (with a shared folder imported into each), dry-run every secrets import, mark plain config with --non-sensitive, use ${KEY} placeholders only where the config loader expands them, switch start scripts to anchor run, and delete the old files only after you confirm |
anchor-ci |
A pipeline, deploy job or server needs secrets | Create a read token scoped to the folders the job reads and pipe it straight into the CI secret, set ANCHOR_API_URL next to ANCHOR_TOKEN, pin the CLI (setup action, install script or npx), and explain the trade-off of --expires-in never |
Every skill keeps secret values out of the agent’s context: values travel from Anchor into your app, or from you into Anchor, never through the conversation. If an agent needs a sensitive value, it gives you the command to run in your own terminal.
For an agent that should read and change secrets directly, with your approval, see Connect AI agents over MCP.
Update or remove
Section titled “Update or remove”npx skills update # update installed skills to the latest versionnpx skills remove anchor-cli # remove one