Grad Student Kunkun Open Sources Method for Managing Inter-Calling Skills

Grad student Kunkun released an open source project that gives teams explicit control over chains of skills that call one another. The approach uses three simple labels, Mermaid diagrams, and a lightweight ask-me skill to keep complex workflows aligned with human intent.

The project addresses a common problem in agent development. When multiple skills interact in sequence or in parallel, models often decide the next step on their own. Kunkun's system moves that decision out of the model and into structured tags and diagrams that a human can audit.

Three-label filtering replaces free-form selection

Kunkun built a small HTML dashboard. Every skill carries three tags: run mode, chain position, and domain. Operators can filter the list instantly and see only the skills that match the current task.

Manual and automatic modes are kept separate. Chain position shows whether a skill runs early, in the middle, or at the end of a sequence. Domain tags group skills by field such as debugging, data cleaning, or documentation. The filters reduce the chance that an unrelated skill is invoked by accident.

Mermaid diagrams make call chains visible

Instead of letting the model improvise the order of calls, the team draws the full chain as a Mermaid flowchart. Each node represents one skill and each arrow shows the data passed to the next step.

When a bug appears or a new requirement arrives, the diagram points directly to the affected group of skills. The same diagram supports pull-request reviews and design changes because the scope of each change stays visible to everyone on the team.

The ask-me skill keeps the model in its lane

Kunkun copied the pattern of the well-known ask Matt skill and created a lighter counterpart called ask me. The skill receives the filtered list, the current Mermaid diagram, and a short summary of the task. It returns only the next recommended skill or asks for human confirmation.

By feeding the model a compressed context rather than the entire library of skills, the system stays within token limits and reduces the risk of unexpected routing. The model no longer needs to guess which skill fits; it receives a short menu that already reflects the team's chosen constraints.

The method avoids two frequent failure modes. First, it prevents the model from selecting skills outside the intended workflow. Second, it keeps the reasoning path legible so engineers can debug or extend the chain without reverse-engineering model decisions.

Kunkun published the full code on GitHub. The repository includes the HTML dashboard, example Mermaid files, and the ask-me skill template. Early users report that the three-label system cuts the time spent on workflow maintenance by roughly half compared with earlier ad-hoc approaches.

The project remains small and focused. It does not claim to solve every agent coordination problem, but it offers a practical pattern for teams that already run chains of skills and want clearer oversight. The repository invites others to adapt the tags and diagrams to their own domains.