Skip to main content
An internal framework for packaging team domain knowledge, workflows, and automation into reusable Claude modules - shareable, versioned, and improvable.

Claude Sauce - Spicey, Yet Nicey

An internal framework for packaging team domain knowledge, workflows, and automation into reusable Claude modules - shareable, versioned, and improvable.

Mor Dabastany By Mor Dabastany • • 689 words • 4 min
ai claude agents marketplace testing agentic-ai

TL;DR

We built an internal framework - Claude Sauce - that lets teams package domain knowledge, workflows, and automation into reusable modules for Claude. Think of it as creating “expert modes” that anyone can use, share, and improve.

The Problem We Set Out to Solve

Claude is extremely capable out of the box - but every organization has its own:

  • Deep domain knowledge no general model can fully capture
  • Established workflows that require specific sequences and checks
  • Tooling and integrations unique to its tech stack
  • Business logic that changes how tasks should be executed

As a result, every specialized use case meant the same friction: people re-explained context, re-wrote instructions, and hoped they didn’t forget critical nuances. Knowledge stayed trapped in individuals instead of becoming reusable.

The Solution: Modular AI Expertise

We created a plugin (or better, a marketplace) that lets anyone package specialized knowledge into shareable, reusable commands, skills, and agents. These turn Claude from a general-purpose assistant into a focused, domain-aware agent - armed with your team’s procedural knowledge.

Instead of prompting from scratch, teams tap into pre-built expertise.

Two Types of The Most Used Items

Skills

Auto-triggered expertise that activates based on context. Example: when working with Jira, Claude already understands your ticket conventions, workflow states, and integration patterns.

(Spoiler: I developed and use it to actually be able to write this article, while Claude is coding for me. Sweet.)

Commands

User-invoked actions via simple slash commands. Example: /deploy-staging runs a predefined deployment workflow, complete with validations and checks.

Why This Matters

1. Knowledge Preservation When an expert creates a skill, their know-how becomes organizational knowledge. New team members get high-quality guidance from day one.

2. Consistency at Scale Instead of 50 people explaining the same process 50 different ways, everyone relies on the same validated workflow.

3. Compound Returns Every new skill benefits everyone. A data team skill helps engineering. A DevOps command helps QA. Value compounds as the library grows.

4. Reduced Context Overhead Claude loads only what’s relevant, exactly when it’s needed. No more copy-pasting walls of instructions.

How to Implement It in Your Organization

I won’t go deep into the conceptual differences between skills, commands, and sub-agents here. What matters is the core idea: each of these capabilities can be packaged, versioned, and distributed as part of a global domain plugin.

You can segment however you like - by skills, commands, sub-agents, or any combination - and describe them declaratively in a manifest file that represents a Claude plugin.

Example Manifest

{
  "name": "internal-ai-modules",
  "owner": {
    "name": "Platform Team",
    "email": "team-mor-dabastany@spicey.internal"
  },
  "metadata": {
    "description": "Shared AI skills and commands",
    "version": "1.0.0"
  },
  "plugins": [
    {
      "name": "project-skills",
      "description": "Common skills for project workflows",
      "source": "./",
      "strict": false,
      "skills": [
        "./skills/issue-tracking",
        "./skills/documentation"
      ]
    },
    {
      "name": "cloud-operations",
      "description": "Cloud-related operational commands",
      "source": "./commands/cloud-operations",
      "strict": false,
      "commands": [
        "./usage-report.md",
        "./enable-logging.md"
      ]
    },
    {
      "name": "test-management",
      "description": "Commands for generating and managing tests",
      "source": "./commands/test-management",
      "strict": false,
      "commands": [
        "./test-generator.md"
      ]
    }
  ]
}

Authoring Skills and Commands

To speed things up, we also use Claude itself to help generate skills and commands. The workflow looks like this:

  1. Maintain a CLAUDE.md file with curated context, conventions, and domain rules.
  2. Feed that context into Claude.
  3. Generate skills, commands, or sub-agents that already conform to your standards.

This creates a tight feedback loop where:

  • Humans define intent and guardrails.
  • Claude helps produce consistent, structured outputs.

Installation & Distribution

The manifest file lives in the project root under:

.claude-plugin/marketplace.json

There are two ways to use it:

/plugin add <repository>

This approach supports a built-in versioning mechanism, making it easier for maintainers to:

  • Release updates
  • Introduce breaking changes intentionally
  • Roll out improvements safely across teams

Option 2: Local Copy

You can also copy the manifest contents directly into the local .claude directory. This works well for experimentation, but doesn’t scale for shared ownership.

And That’s It - You Are Good to Go!

If your organization is ready to convert team expertise into reusable Claude modules, start small, share often, and watch how workflows multiply in efficiency.

“LET’S MAKE AI GREAT - AGAIN!”

Related Posts