Skip to content

BMS Skills

Skills are reusable instruction sets that extend AI coding agents with team-specific workflows, conventions, and domain knowledge.

Instead of explaining your stack to the agent every session, you install a skill once — and it activates automatically whenever it's relevant.

What a skill does

A skill is a SKILL.md file that tells the agent:

  • What conventions to follow — e.g. use uv not pip, use ty check not pyright, no ORM
  • What patterns to apply — e.g. asyncpg t-string SQL, TanStack query key factories, FastAPI lifespan pools
  • What to watch out for — common pitfalls, anti-patterns, project-specific constraints

Skills activate in two ways:

1. Slash command — explicitly invoke any skill: 

/coding-guidelines-python
/spark-connect
/init-app-stack

2. Auto-trigger — the agent reads the skill description and loads it automatically when the context matches. You don't have to do anything — if you're writing FastAPI code and the fastapi-guideline skill is installed, the agent picks it up.

What's included

Skills are grouped into plugins by domain:

Plugin What's inside
coding Python guidelines, TypeScript guidelines, TanStack, FastAPI, Azure auth, Postgres, autoresearch, init-app-stack scaffold
databricks Databricks CLI, Spark Connect version alignment
ui KendoReact components
writing Remove AI slop, scientific revision
bms Master skill — all core standards via /bms
caveman Ultra-compressed communication (~75% fewer tokens)
azure-deploy Azure deployments via azd / terraform
azure-diagnostics Azure production diagnostics

→ Install now

Automatic tool remapping

The coding and bms plugins ship with PreToolUse hooks that silently rewrite shell commands before the agent runs them:

You type Agent runs
python script.py uv run python script.py
pip install foo uv add foo
pytest uv run pytest
npm install foo bun add foo
npx foo bun x foo

No config needed — hooks activate automatically when a plugin is installed.

We highly recommend Caveman by Julius Brussee. It cuts token usage ~75% by having the agent communicate in ultra-compressed style while keeping full technical accuracy. Less noise, faster responses, lower cost.

"caveman@bmsuisse-skills": true

Or install directly:

bun x skills add https://github.com/JuliusBrussee/caveman

Full credit to Julius Brussee for creating and maintaining Caveman. Outstanding work.