Intern in context engineering. Build public software.
RobOS is built by people steering coding agents. Contributing here means learning to brief an agent, give it the right knowledge, and prove what it did, on a real Apache-2.0 codebase where every merged pull request carries your name.
No application, no fixed dates. Go at your own pace.
Why intern here
The whole project is a context-engineering exercise
Context engineering is deciding what an agent knows, when it knows it, and how you check its work. Most people learn it on private projects nobody else sees. In RobOS the instructions, skills, knowledge graph and specs are all in the open, and improving them is the work.
Your work is public
Commits, reviews and discussions live on GitHub under your name. A merged PR is a better reference than a certificate.
Agents type, you think
You aren't expected to hand-write every line. You frame the task, give the agent the right context, and prove the result. That's the skill teams hire for now.
Feedback is built in
Every PR ships with test output and, for anything visible, a recorded walkthrough. Reviewers comment on how you scoped and verified the work, not just the diff.
What you'll practise
Six skills, each with a home in the repo
| Skill | Where you practise it |
|---|---|
| Writing agent instructions | AGENTS.md and the per-tool files. Tighten a rule and watch agents behave differently. |
| Authoring skills | .agents/skills/: short SKILL.md playbooks agents run, like create-robos-app and kgraph-diff. |
| Modelling knowledge for agents | .robos/kgraphs/: JSON-LD checked against SHACL shapes, so an agent gets only the slice it needs. |
| Writing specs as prompt contracts | docs/ideas/specs/ and docs/issues/reported/: tasks any contributor's agent can pick up and finish. |
| Verifying agent output | E2E tests, the headless container run and narrated recordings. Prove the work instead of trusting it. |
| Reviewing agent pull requests | Read other contributors' PRs, check their evidence and leave useful comments. |
Pick a track
Choose a focus for your first few PRs
Instructions and skills
Fix a skill agents misuse, or write one for a task you keep repeating. Start by running a skill end to end and fixing whatever confused the agent.
Knowledge graph
Add or correct entities, validate them, and check the blast radius with kgraph-diff. Start with an app whose graph entry is out of date.
Specs and issues
Turn a bug or idea into a task another contributor's agent can run. Start by reproducing a reported issue and sharpening its steps.
Apps and tests
Fix a bug or build a feature in one of the Electron apps, with a test and a recording as proof. Start with an issue that touches one app.
Your first week
From clone to merged PR
- Read the contextAGENTS.md, a few skills, one graph package
- Run somethingLaunch an app and the unit tests
- Say helloTell us your track in Discord
- Ship one small PRA doc fix, a sharper skill, a graph node
- Fix what tripped youMake it easier for the next person
git clone https://github.com/nddipiazza/robos.git
cd robos
electron packages/dev-central
cd packages/robos-test && npm install && npm run test:unitNeeds Node.js 20+ and a coding agent. Docker is used for the headless test run. Each package has its own dependencies; there's no root package.json.
How a contribution flows
You steer, the agent builds, the evidence decides
- Pick a taskAn issue, a spec, or your own idea written up
- Brief the agentAGENTS.md, the spec and the right skills
- ImplementOn a feature branch, with you reading along
- Prove itTests plus a recorded walkthrough
- Open a PREvidence attached, and a note on how you briefed the agent
The contributor guide has copy-paste prompts for fixing an issue, implementing a spec, creating an app, improving a skill and updating the knowledge graph, along with the conventions reviewers check.