dk-ai

Custom AI agents and skills for building software with dk and the dk0 build tool. They teach a coding agent (Claude, Copilot, Cursor, and similar) to analyze dk projects, create and repair dk packages, and turn a plain-language software request into a build plan that ships through dk0 remote.

The given-user-prompt-make-new-prompt-that-uses-dk agent is also the system prompt behind the dk Prompt Studio prompt bar on https://diskuv.com (it runs there on Amazon Nova Micro). The mini-plans the Studio produces tell a user's agent to install these skills, so this repository is where that install points.

Layout

• agents/<name>.agent.md - task agents (see the live inventory in AGENTS.md).
• skills/<name>/SKILL.md - reusable skills the agents (and you) invoke.
• tests/ - per-agent and per-skill tests.
• AGENTS.md - the authoring conventions and the current inventory of what is here.

Install

The files use the Claude agent/skill format (agents/*.agent.md, skills/*/SKILL.md). Exact discovery rules vary by tool and version, so pick the closest path below; when in doubt, the "other agents" path always works.

Claude Code

Clone the repo and copy its agents and skills where Claude Code looks for them, either for one project or for every project:

git clone https://github.com/diskuv/dk-ai
# this project only:
mkdir -p .claude/agents .claude/skills
cp -r dk-ai/agents/* .claude/agents/
cp -r dk-ai/skills/* .claude/skills/
# or for every project (user-wide):
mkdir -p ~/.claude/agents ~/.claude/skills
cp -r dk-ai/agents/* ~/.claude/agents/
cp -r dk-ai/skills/* ~/.claude/skills/

Reload Claude Code so it discovers them.

Claude Desktop / claude.ai

Add each folder under skills/ through the Skills feature. For an agent, paste the body of the relevant agents/*.agent.md as your project instructions.

Cursor, GitHub Copilot, OpenAI, and other agents

There is no cross-tool standard for installing skills, so point your agent at the files directly. The most reliable way is to clone the repo into your project and ask your agent to read the skill or agent it needs:

git clone https://github.com/diskuv/dk-ai
# then tell your agent: "read dk-ai/skills/analyze-dk-project/SKILL.md and follow it"

If your agent has a web or fetch tool, it can pull a skill into context without cloning:

https://raw.githubusercontent.com/diskuv/dk-ai/main/skills/<name>/SKILL.md
https://raw.githubusercontent.com/diskuv/dk-ai/main/agents/<name>.agent.md

This three-step fallback (install the skills, else fetch them with a web tool, else clone and read them) is exactly what the dk Prompt Studio mini-plans tell a user's agent to do, so any of these paths satisfies a generated mini-plan.

What is here

See AGENTS.md for the self-maintained inventory of every agent and skill, each with a one-line "use when" and its hard-stop rules.