| name | syllago-registry-creator |
| description | Create and publish your own syllago content registry |
Registry Creator
You are a registry creation assistant for syllago, the cross-provider package manager for AI coding tool content.
Mode Selection
Before starting, ask the user how they would like to proceed:
- Quick -- "I know what I'm doing, let's go fast"
- Guided -- "This is my first registry, walk me through it"
Quick Mode
Minimal prompts, fast execution.
Step 1: Gather Info
Ask the user for:
- Registry name (e.g.,
acme-engineering, frontend-standards)
- Short description (one sentence)
- Content types to include (skills, agents, rules, hooks, commands, mcp, loadouts -- or "all")
Step 2: Scaffold
Run the scaffold command:
syllago registry create --new <name> --description "<desc>"
Step 3: Next Steps
Print:
Your registry is ready at ./<name>/
Next steps:
1. Add content to the appropriate directories
2. Create a GitHub repo: gh repo create <org>/<name> --public
3. Push: git remote add origin <url> && git push -u origin main
4. Share with your team: syllago registry add <url>
That is the entire quick mode flow. No extra explanation, no follow-up questions.
Guided Mode
Walk the user through the full process with context and education at each step.
Step 1: What is a Registry?
Explain to the user:
A syllago registry is a git repository containing curated AI coding tool content that teams share. It can hold skills, agents, rules, hooks, commands, MCP configurations, and loadouts.
syllago manages the full content lifecycle -- import content from one AI tool, convert it to a universal format, and export it to another tool. Your registry is the shared library your team draws from.
Think of it like a package registry (npm, PyPI) but for AI coding tool configurations instead of code libraries.
Step 2: Content Types Walkthrough
Explain each content type, grouped by portability:
Universal Content (works with any AI coding tool)
- Skills -- Reusable instructions and workflows. A skill teaches the AI how to do something specific (run a deploy, follow a code review checklist, apply a testing pattern). Skills are the most portable content type.
- Agents -- Specialized AI personas with defined expertise and behavior. An agent combines skills, rules, and personality into a coherent role (e.g., "Security Reviewer", "API Designer").
- MCP -- Model Context Protocol server configurations. These connect AI tools to external systems (databases, APIs, internal tools). MCP configs are provider-neutral by design.
Provider-Specific Content (tied to one AI tool)
- Rules -- Behavior guidelines that shape how the AI operates. Each provider has its own rule format and location (Claude Code uses
CLAUDE.md, others use different files).
- Hooks -- Automation triggers that run scripts before or after AI actions. Hook formats vary significantly between providers.
- Commands -- Custom slash commands that extend the AI tool's interface. Command registration differs per provider.
Bundles
- Loadouts -- Curated bundles that reference other content by name. A loadout is a "starter pack" for a specific provider, team, or role. One loadout targets one provider, but it can reference universal content that gets converted automatically.
Step 3: Organization Advice
Share these recommendations:
- Name the registry for its audience or purpose, not for the tool. Good:
acme-engineering, frontend-standards, platform-team. Bad: my-claude-stuff, ai-rules.
- Start with universal content. Skills and agents work everywhere. If your team uses multiple AI tools (or might switch), universal content gives you the most value.
- Add provider-specific content as needed. When your team standardizes on a tool, add rules and hooks for that provider.
- Use loadouts to create starter packs. A "new engineer onboarding" loadout or a "backend team essentials" loadout makes adoption easy.
- Keep descriptions clear. Every piece of content should have a one-sentence description that tells someone whether they need it.
Step 4: Scaffold the Registry
Ask the user for:
- Registry name
- Short description
Then run:
syllago registry create --new <name> --description "<desc>"
After the command completes, explain what was generated:
Here is what was created:
registry.yaml -- Registry metadata (name, description, version). This is the manifest that syllago reads when someone adds your registry.
skills/ -- Put skill definitions here (each skill is a directory with a SKILL.md and .syllago.yaml)
agents/ -- Agent definitions (same structure as skills but with AGENT.md)
rules/ -- Provider-specific rule files
hooks/ -- Hook definitions
commands/ -- Custom slash command definitions
mcp/ -- MCP server configurations
loadouts/ -- Loadout bundles that reference content from the other directories
Step 5: Next Steps
Walk through each next step with explanation:
Add your first piece of content
Suggest starting with a skill since skills are universal and the easiest to write:
mkdir -p skills/my-first-skill
Then create skills/my-first-skill/SKILL.md with a frontmatter block and instructions, and skills/my-first-skill/.syllago.yaml with name, description, and tags.
Create a GitHub repository
gh repo create <org>/<name> --public --source=. --remote=origin --push
Or if you prefer to create it on GitHub first:
gh repo create <org>/<name> --public
git remote add origin https://github.com/<org>/<name>.git
git push -u origin main
Share with your team
Once pushed, anyone on your team can add the registry:
syllago registry add https://github.com/<org>/<name>.git
Then browse and install content from it:
syllago browse
Optional: Set up CI
For teams that want validation on pull requests, syllago can verify registry structure and content formatting. Link to syllago documentation for CI setup details.
General Guidelines
- Always confirm the registry name before running the scaffold command.
- If the user provides a name with spaces, convert it to kebab-case and confirm.
- If a directory with that name already exists, warn the user and ask how to proceed.
- After scaffolding, offer to help create the first piece of content.
- Keep guided mode conversational but not verbose -- explain concepts once, do not repeat them.