Using a coding agent¶

You can use an AI coding agent (GitHub Copilot in VS Code/Codespaces, Claude Code, OpenAI Codex, Cursor and others) as a lab partner. The repository ships guidance so your agent understands this environment:

File	Read by	Contains
`AGENTS.md`	Copilot, Codex, Cursor, Gemini CLI, most agents	How KID is wired, safety rules, where things live, how to validate changes
`CLAUDE.md`	Claude Code	Imports `AGENTS.md`
`.agents/skills/`	Agents that support Agent Skills (also linked as `.claude/skills` and `.github/skills`)	Step-by-step procedures, loaded only when relevant

Skills¶

Skill	Ask things like...
`kid-lab-guide`	"I'm on lab 3 step 2, why did my connection time out?" · "Give me a hint for lab 6 question 3"
`kid-network-policy`	"Let project beta install packages from PyPI" · "Show me what alpha's workspace tried to reach"
`kid-new-project`	"Create project gamma with medium workspaces and give researcher2 access"
`kid-kyverno-policy`	"Why was my pod rejected?" · "Write an Audit policy that requires a cost-centre label"
`kid-workspace-config`	"Make idle workspaces stop after 10 minutes" · "Add a large size to project beta"
`kid-troubleshoot`	"Argo CD says unable to resolve revision" · "My workspace won't start"

Tips¶

Run the agent inside the dev container (the default in Codespaces and VS Code Dev Containers). That's where kubectl, hubble and friends can reach the cluster.
The guidance tells the agent to teach rather than just solve: it gives hints for the "Check your understanding" questions unless you ask for the answer. Ask it to explain its commands.
The agent is told not to weaken guardrails (network policies, Kyverno, quotas) to make something work, and to ask before destructive actions. If a lab is about breaking something, say so.
Changes under gitops/ only reach the cluster after you commit and push to your fork (then tre sync). The agent should tell you which changes are GitOps and which are direct kubectl.
Treat agent output like a colleague's: read the diff before you commit it.