Documentation menu
Reference

The Skills plugin

Skills run on top of the Platform. They are markdown prompts — a Claude Code plugin — that orchestrate the knowledge base into coherent workflows: planning a feature, running tests, cutting a release, handling an incident. Each skill drives the Platform's MCP tools, so it works against the knowledge base on its own — no extra services required.

Kvendra Skills (Apache-2.0) is a Claude Code plugin (markdown prompts; no runtime code).

The official Claude Code plugin that materializes the skills layer: the single set of skills for the project, distributed as a markdown plugin through a public marketplace. It is the join point that orchestrates the CLI and the KB.

Most skills run on the Platform alone — planning, drafting entities in a transaction, assembling a changelog are pure knowledge-base operations. The optional CLI broker is required only by skills that perform sensitive system operations (a git push, a deploy), where it routes the operation so the language model never sees a raw token or key. You can fork the open-source skills to change which steps use the broker.

Direction

  • A single set of skills for Claude Code (the parallel community track is deprecated).
  • English as the source language; user-facing output follows the language declared in the project config.
  • Thin, KB/STD-driven skills: tech-specific recipes live in KB STD entities, not hardcoded in the skill markdown.

The skill catalogue

  • Pipelines: new-feature, bug, release-manager, incident-manager, regression.
  • Subagents: planner, implementer, validator, tester, analyzer, updater, interface-validator.
  • Documentation: doc-validator.
  • Setup & meta: setup, env-check, consultancy, to-do, to-do-summary, requirements-analyst, changelog, user-help.
  • Deploy: deploy.

LLM target: Any LLM that Claude Code supports.

Thin skills, knowledge in the KB

A skill does not hardcode the specifics of your stack. Instead, the reusable recipes live as standard/playbook entities (STD) in the knowledge base, and the skill reads them at runtime. That keeps the prompts small and portable while letting each project carry its own conventions in a place that is versioned and auditable.

Skills run on the Platform — the CLI is optional

A skill talks to the Platform through its MCP tools — the same fourteen that read and write the knowledge base. That is all most skills need, so the majority run with just the Platform: no CLI, no broker. Planning a feature, drafting entities inside a transaction, or assembling a release changelog are pure knowledge-base operations.

The optional CLI broker enters only for the skills that perform system operations — pushing to a git remote, calling a cloud API, running a shell command, as a deploy skill would. For those, the operation is routed through the broker so the model orchestrating the work never sees a raw token or key. Because the skills are open source, you can read exactly which steps need the broker and adapt them.