Building Composable Skills for Claude Subagents

Claude skills/commands/subagents composition is important: - create a generic skill like "nano-banana" with a script to use the Gemini API for image gen - create a generic skill like "invert-image" that can create dark mode variants of any image (for example) - create a generic skill like "upload-image-to-notion" that has access to the API - create a generic skill like "write-guide-from-codebase" that can analyze a part of your codebase and write a tutorial for how it works — can include visualizations, diagrams, code snippets, etc. (bonus points: connect it to Notion MCP or wherever you host your docs so it can publish the guide automatically) Then compose all of these into a subagent like: `guide-writer` — the subagent knows how to analyze the codebase, write the guide, come up with the best prompt for an image, gen the image, make the dark mode variant, create a Notion page for the guide, and upload the images to the guide. The subagent works and keeps the main context thread clean. You only supply the initial prompt for the guide you want to write, and the main agent will respond when it's done with a link to the guide. The important thing is because it's all well-composed, you can use the skills individually at any point, or easily create new subagents that use the same skills for entirely different workflows.