fine-tuning-expert

Get started in minutes with our [Quick Start Guide](QUICKSTART.md).

TL;DR:

```bash

/plugin marketplace add jeffallan/claude-skills

```

then

```bash

/plugin install fullstack-dev-skills@jeffallan

```

> New: Use /common-ground to surface and validate Claude's assumptions about your project before starting work. Add --graph to visualize the reasoning structure as a mermaid diagram.

> Note: The [Project Workflow Commands](#project-workflow-commands) require an Atlassian MCP server for Jira and Confluence integration. See the [Atlassian MCP Setup Guide](docs/ATLASSIAN_MCP_SETUP.md) for configuration instructions.

Progressive Disclosure Pattern

Each skill follows this structure:

```

skills/react-expert/

├── SKILL.md # Lean core (80 lines)

│ ├── Role definition

│ ├── When to use

│ ├── Core workflow

│ └── Routing table # Points to references

└── references/ # Loaded on-demand

├── server-components.md # RSC patterns

├── react-19-features.md # use() hook, actions

├── state-management.md # Context, Zustand, Redux

├── hooks-patterns.md # Custom hooks, optimization

├── performance.md # memo, lazy, virtualization

└── testing-react.md # Testing Library patterns

```

How It Works:

1. Skill loads with minimal context (~80 lines)
2. Claude reads the routing table
3. Loads specific references only when context requires
4. 50% faster initial responses, surgical precision when needed

Project Structure

```

claude-skills/

├── .claude-plugin/

│ ├── plugin.json # Plugin metadata

│ └── marketplace.json # Marketplace configuration

├── skills/ # 65 specialized skills

│ ├── react-expert/

│ │