Agent skills for installing, configuring, and operating Coder from the command line.
| Skill | What it does |
|---|---|
setup |
End-to-end install and first-run setup of a Coder deployment without the web UI: install, sign in the first admin (GitHub device flow or generated email/password), push a starter template, optionally build the first workspace. Covers Docker, Kubernetes, and direct host installs. |
The skills CLI works with
every major coding-agent CLI. It resolves a marketplace, picks a
skill, and drops it under each agent's discovery directory:
# project-local: drops the skill in ./.claude/skills, ./.agents/skills, ...
npx skills add coder/skills@setup
# user-global: drops it in ~/.claude/skills, ~/.codex/skills, ~/.cursor/skills, ...
npx skills add coder/skills@setup --global
# install only for a specific agent
npx skills add coder/skills@setup --agent codex
npx skills add coder/skills@setup --agent claude-code
npx skills add coder/skills@setup --agent cursorOnce installed, the skill activates automatically when you ask the agent to install or set up Coder. There is no slash command; just say what you want ("set up Coder locally", "deploy Coder behind HTTPS", "stand up Coder on Kubernetes") and the agent picks the skill from its description.
Claude Code can also load this repo as a plugin marketplace directly:
/plugin marketplace add coder/skills
/plugin install coder@coder-skills
If the install doesn't take effect immediately, run
/reload-plugins and try again.
For local development, point the marketplace at a checkout:
/plugin marketplace add /path/to/skills
In headless mode (claude -p), use --plugin-dir instead:
claude -p \
--plugin-dir /path/to/skills \
--permission-mode bypassPermissions \
"Use the setup skill to install Coder via Docker compose, \
bootstrap an admin user, push the docker starter template, and \
create one workspace named demo."--permission-mode bypassPermissions is required because the skill
runs curl | sh, starts background processes, and writes config
files. Only use it on a sandboxed host.
The Codex CLI has its own plugin marketplace command. Same source:
codex plugin marketplace add coder/skillsCodex auto-discovers any skill placed under $CODEX_HOME/skills
(default ~/.codex/skills), so the npx skills add coder/skills --agent codex path above works without any extra step.
Two test harnesses ship with the repo, one per CLI. Both spin a real Coder server in Docker, drive the skill end-to-end, and verify the result via the Coder REST API.
./test/run.sh # Claude Code (claude -p)
./test/run-codex.sh # Codex (codex exec)See test/README.md for what each harness
expects and how to point them at a different port or template.
Each skill is a directory under skills/<name>/ containing a
SKILL.md with YAML frontmatter, plus optional references/,
scripts/, and assets/ subdirectories. The frontmatter requires:
---
name: kebab-case-skill-name
description: One sentence on what it does and exactly when to use it.
---Anthropic's skill spec is at https://github.com/anthropics/skills/tree/main/spec. Conventions this repo follows on top of that:
- Keep
descriptionunder 1024 characters. Codex truncates the description at that limit and drops the skill if the truncated YAML is invalid. Claude tolerates longer descriptions but does not benefit from them. Push activation triggers and exclusions into a tight 1000-char block. - Keep
SKILL.mdlean (target 1500 to 2500 words). Push detailed matrices, tables, and edge-case lists intoreferences/*.mdand lazy-load them by trigger. - Reference files from
SKILL.mdwith relative paths so the skill works at the personal, project, or plugin level and across agents. - Don't ship secrets, license keys, or auth tokens in any skill artifact. Skills are public.
- Add a smoke test under
test/that runs the skill headlessly and verifies the externally observable result. The two existing harnesses (test/run.sh,test/run-codex.sh) are good templates.
To be set by the publisher. Until then, treat as "all rights reserved".