Skip to content

Add a kit skill command to teach AI agents how to use the kit CLI #1249

Description

@gorkem

Problem

Coding agents (Claude Code, Codex, Cursor, Copilot, Gemini CLI) increasingly drive CLIs directly. Today an agent's only way to learn kit is --help output, which is flag-oriented and says nothing about workflows: Kitfile authoring, pack/push/pull/unpack, tagging. Users who want agents working with ModelKits hand-write AGENTS.md sections or paste docs into context, and those drift from the installed kit version.

Other agent-facing CLIs have converged on shipping this knowledge as an installable skill following the Agent Skills spec:

  • hunk ships skills/hunk-review/SKILL.md in-repo and provides a hunk skill command to locate/load it
  • beads has bd setup <agent> to install a skill and agent instructions, plus bd prime to print workflow context on demand

Proposal

Add a kit skill command that installs a SKILL.md (covering Kitfile authoring, the pack/push/pull/unpack workflow, tagging, and common errors) into the appropriate agent directory (.claude/skills/, the shared .agents/skills/, etc.).

The skill should be embedded in the binary so it always matches the installed kit version and works air-gapped.

We should also publish the skill as a ModelKit. It's the natural distribution channel for our own project: the skill becomes a versioned, verifiable OCI artifact that agents and CI can pull like any other ModelKit, and it demonstrates skills-as-artifacts with our own tooling.

Related

kit unpack --as-skill: unpacking a ModelKit directly into an agent skills directory. The kit CLI skill published as a ModelKit would be a natural first consumer of that path.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Fields

    No fields configured for Feature.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions