Skills
ERun ships a small set of skills — guidance bundles the Agent loads on demand to do project-shaped work well. You don't install them piece by piece; opening an env gives you the in-pod set, and a single command gives you the same set on your laptop. Then you just describe what you want — the Agent picks the matching skill and writes the code by hand, idiomatic for your project.
For the conceptual model see Concepts · Skills. For the SKILL.md format and the plugin manifest schemas see Skills spec.
Inside a deployed env
Nothing to do. erun open brings up an env whose runtime image already has the skills baked in; on env start the entrypoint installs each one into the Agent's discovery directory (~/.claude/skills/<name>/ for Claude Code, ~/.codex/skills/<name>/ for Codex). The Agent picks them up on its next prompt.
Edits you make to a skill file inside the env survive pod restarts. Only a fresh home (new env, new tenant) re-pulls the baked copy.
On your laptop
Install the ERun plugin once, and the same skills load into your local Claude Code alongside whatever else you have:
/plugin marketplace add sophium/erun
/plugin install erun-tools@sophium/erun
Skills become invocable as /erun-tools:<skill-name>. To update later: /plugin marketplace update sophium/erun.
Codex doesn't have an analogous plugin marketplace yet. Inside a deployed env Codex gets the skills automatically; on a laptop, copy the SKILL.md files into ~/.codex/skills/<name>/ manually until upstream Codex ships plugin support.
What's in the set
Skills come in two kinds.
Blueprint skills
Package ERun's accumulated best practices for building complex industry-strength solutions.
| Skill | What it does | Trigger by saying |
|---|---|---|
erun-blueprint-rls-db | Build a multi-tenant PostgreSQL schema with row-level security, Atlas migrations, UUIDv7 keys, and the canonical tenant/issuer/user bootstrap. | "build a multi-tenant postgres database", "create a tenant-scoped postgres schema with row-level security" |
erun-blueprint-api | Build a multi-tenant Go HTTP API with OIDC bearer auth, tenant-from-issuer resolution, layered model / repository / routes, transaction-scoped RLS context, and audit logging. | "build a multi-tenant http api", "create an erun-backend-api-shaped service" |
Workflow skills
Let you participate in ERun's processes — report problems, share improvements back so other users benefit.
| Skill | What it does | Trigger by saying |
|---|---|---|
erun-file-issue | File a bug or feature against ERun on GitHub. | "file an erun bug", "register erun feature", "open an erun issue" |
erun-contribute | Create a new issue against sophium/erun, then drive the full clone → branch → implement → PR motion to share your improvement back. | "contribute to erun", "submit a PR to erun", "propose an improvement to erun" |
erun-build-env | Extend ERun's published runtime image with your project's own toolchain and point the environment at the custom image. | "init build environment", "create a custom build environment", "customize the runtime image" |
erun-pin-version | Change or try the erun version an environment uses, re-pinning every place it is recorded in one verified, idempotent motion — and revert just as easily. | "change the erun version", "pin erun to 1.0.174", "revert the erun version" |
You don't need to memorise the trigger phrases — the Agent matches against the skill's description field, which lists several variants. If you describe the work in plain language ("I want to add a Postgres database that's tenant-scoped"), the right skill usually fires on its own.
User-authored skills
You can layer your own guidance on top — house style, framework preferences, audit rules — by adding a SKILL.md under <repo>/.erun/skills/<name>/. (Planned: built-in support; today the runtime image only loads the baked set.) For the format and the layering precedence rules see Skills spec.
Where next
- Build a small app — watch the Agent use the in-pod skill set to add a service.
- Concepts · Skills — the conceptual model.
- Skills spec — SKILL.md format, marketplace manifest, error behaviour.