Docs · Agents

Skills reference.

A skill is a markdown file with frontmatter that teaches an agent how to operate part of the stack. Remoco workstations ship with a skills pack baked in; you can layer your own on top.

Anatomy of a skill

---
name: remoco-artifact
description: Create and publish an HTML artifact through remoco
trigger: "publish an artifact | create a remoco artifact | remoco artifact push"
---

# Markdown body: explain when to use this skill, what commands to run,
# what the output looks like, what to return to the user.

Where skills live

On agent session start, remoco-skills-sync concatenates all sources in that priority order and prepends them to the harness context.

Skills shipped with remoco

NameTriggerPurpose
remoco"operate the workstation"Core ops guide: SSH, tmux, expose, worktree, suspend/resume.
remoco-artifact"publish an artifact"Create + push HTML reports via the control plane. Returns public URL.
remoco-ds"use the design system"Load a custom DS (tokens + primitives) for artifact generation.
remoco-debugger"inspect the workstation"Read the debugger dashboard at :9999 over HTTP.

Writing your own

Drop a new file at ~/.remoco/skills/my-skill.md:

---
name: my-repo
description: Operate the my-repo codebase on this workstation
trigger: "fix a bug in my-repo | deploy my-repo"
---

# my-repo

Repo lives at ~/code/my-repo.
Start services with make dev.
Tests: pytest tests/.
Service port: :8090 → alias api.main.<me>.remoco.dev.

Then reload:

$ remoco resync
# skills-sync indexed 4 baked + 1 user + 0 repo = 5 skills

Design defaults (for artifact skills)

Artifact-generating skills should link the default Geist dark tokens unless a custom DS is loaded:

<link rel="stylesheet" href="https://storage.googleapis.com/remoco-artifacts/ds/geist-dark/tokens.css">

Then include the standard eyebrow + H1 + meta row at the top, and a Centari-style provenance stamp at the bottom. Full baseline at gs://remoco-artifacts/ds/geist-dark/primitives/baseline.html.

Skills are dumb markdown. No runtime, no execution. The agent reads them as context and decides what to do. If a skill is wrong, edit the file.

← CLI reference Back to docs →