Prompts & Commands
Rich prompts and ready-to-use slash commands that load instructions for AI agents.
Read this when you want to understand how Catalyst commands work, or when looking for the right command for your task.
Useful for anyone working with AI agents in a Catalyst project.
What Are Slash Commands?
Slash commands are shortcuts that load instructions for AI agents. For example, type /code in your chat, and the AI loads the coding session setup — reading project context, following conventions, and setting guardrails.
Commands work in both Claude Code and VS Code Copilot. They load prompt files from /.catalyst/prompts/, giving agents consistent behaviour across sessions.
How They Work
Commands are thin wrappers that reference detailed prompts:
You type a slash command
You run /code in your AI chat
The command loads a prompt file
Points to .catalyst/prompts/repo-code.md (and related files)
The AI receives instructions
Knows to read project context, follow AGENTS.md, and set up properly
You continue the conversation
The AI now has the right context and constraints for the task
Six Command Groups
Commands are organized by purpose — pick the group that matches your task:
Chat
Start project conversations, continue work, and create handoff summaries.
/chat/continue/summaryDelivery
Run the Brief → Build → Review → Refine cycle that keeps projects aligned.
/brief/build/review/refineProject
Plan features, design visual identity, and validate alignment with vision.
/plan/design/alignArtefacts
Create and update the four core project specs that capture intent.
/create-vision/create-experience/create-brand/create-architectureSupport
Get recommendations, think through problems, and make decisions with trade-offs.
/next/think/decide/debugRepo
Set up projects, run audits, prepare releases, and manage stage advancement.
/code/setup/status/audit/harden/release/promote/helpWhy This Matters
Without commands, every AI session starts from scratch. You explain conventions, reference docs, and hope the AI follows them. With commands, the AI loads everything it needs in one step — making sessions consistent, faster, and more reliable.
Commands also separate what to do (the command) from how to do it (the prompt). This means you can update prompts once and every tool that uses them gets the improvement.
Common Workflows
Here's how commands fit into typical tasks:
Starting a coding session
/codeLoads project context, conventions, and coding guardrails
Building a new feature
/brief/build/review/refineRun the delivery loop to build, validate, and iterate
Setting up a new project
/setup/create-vision/create-architectureInitialize repo and create foundational specs
Stuck or unsure what to do
/next/think/decideGet recommendations or work through decisions
Preparing for launch
/audit/harden/release/promoteCheck quality, fix issues, prepare release, advance stage
Commands make AI agents consistent. Instead of hoping the AI remembers conventions from the last session, commands load everything it needs — context, constraints, and working style — in one step. This turns AI from a helpful assistant into a reliable team member.
Next Steps
Pick a command group to explore: