Docker Agent can now load skills straight from a GitHub repo
v1.149.0 lets a YAML-defined agent reference a public GitHub repo as a skill source, so instructions no longer have to live in the same project or be copied by hand.
Docker Agent, the CLI plugin for defining and running AI agents in YAML, shipped v1.149.0 on 7 October. The headline change: a skills: entry can now point straight at a public GitHub repository URL, and the agent pulls the skill from there instead of needing it copied into the project.
A skill is a set of instructions an agent loads on demand when a task matches its description, rather than being stuffed into the system prompt up front. Before this release, that meant the skill had to live locally. Now a repo URL works directly, with query parameters to pin a branch, tag, commit or subdirectory.
agents:
root:
model: openai/gpt-4o-mini
skills:
- local
- https://github.com/docker/skills
- https://github.com/owner/repo?ref=v0.3.0&path=skillsOnly public GitHub repos are supported, with an optional GITHUB_TOKEN to avoid rate limits.
Why it matters: skills defined this way can be versioned and shared across teams or projects without copy-pasting instruction files into every agent config. It’s a small change, but it’s the kind of distribution mechanism that decides whether a skills ecosystem catches on or stays bespoke to each repo.