Out Plane Agent Skills

Teaches a coding agent to deploy and operate applications on Out Plane: applications and deployments, environment variables, ports, custom domains, volumes, managed PostgreSQL, IP access, logs and metrics.

One skill, use-outplane. It loads on its own when the conversation is about Out Plane, so there is nothing to remember and nothing to type.

Install

With the Out Plane CLI (needs CLI v0.2.1 or later):

outplane skills install

Detects every coding tool on the machine and installs into each. --agent picks one, --project writes into the current repository instead of your home directory so the team shares one copy.

As a Claude Code plugin:

/plugin marketplace add outplane/skills
/plugin install outplane@outplane

With the skills CLI:

npx skills add outplane/skills

By hand. Copy skills/use-outplane/ into your agent's skills directory:

Agent	Directory
Claude Code	~/.claude/skills/
Cursor	~/.cursor/skills/
OpenAI Codex	~/.codex/skills/
OpenCode	~/.config/opencode/skills/

Restart the agent afterwards so it picks the skill up.

What it needs

The outplane CLI, signed in:

curl -fsSL https://outplane.com/install.sh | sh   # macOS, Linux
npm i -g outplane                                  # Windows, or wherever Node is present

outplane login

In a pipeline, set OUTPLANE_TOKEN instead of signing in.

CLI source and releases: github.com/outplane/cli. Documentation: docs.outplane.com/cli.

Try it

Once it is installed, ask your agent things like:

• "Deploy this directory to Out Plane."
• "Why did my last deploy fail?"
• "Set DATABASE_URL on checkout and roll it out."
• "Give this app a custom domain."
• "Is my app healthy? What are the 5xx responses?"
• "Create a Postgres database and wire it into the app."
• "Only let the office network reach this service."

How it is built

skills/use-outplane/
├── SKILL.md              routing, execution rules, the destructive protocol
└── references/           depth, loaded only when a task needs it
    ├── deploy.md         deploy, build
    ├── configure.md      env, env group, port
    ├── network.md        domain, ip-profile
    ├── data.md           db, volume
    ├── operate.md        logs, requests, metrics, shell, instances, scale, pause
    ├── manage.md         app lifecycle, registry credentials
    ├── session.md        login, link, team, status, update, api
    ├── errors.md         exit codes, the error object, every error code
    └── ci.md             running where there is no terminal

SKILL.md stays small on purpose. Once a skill loads, its content stays in the conversation for the rest of the session, so the body is routing and the depth waits in references/ until a task actually needs it.

Nothing here restates the CLI. Command lists, flags, allowed values and per-command error codes are not copied into these files, because a copy is wrong the first time the CLI changes and stays wrong until somebody notices. The skill sends the agent to outplane schema, which is local, needs no credential and is generated from what the binary actually runs. Only the exit-code contract is written out, because it is append-only by design and an agent branching on it should not need a tool call.

node scripts/validate.mjs          # check the rules a skill has to follow

Contributing

Read the skill before changing it. A reference exists for the recipes: which commands compose into a task, in what order, and what to check between the steps. Flag lists, allowed values and limits belong in outplane schema and in the documentation, not here.

Two rules that are not negotiable:

1. Bump version in .claude-plugin/plugin.json in any change to skill content. Claude Code compares that version to decide whether an installed copy is stale. Without a bump, nobody receives the change. Keep metadata.version in SKILL.md in step with it; scripts/validate.mjs checks that they agree.
2. node scripts/validate.mjs must pass. It runs in CI.

License

Apache-2.0. Copyright 2026 Out Plane LLC.