Make coding agents great at using your API
summary
Claude Code already makes ~4% of measurable GitHub commits and is forecasted to reach ~20% by end of 2026, so the practical work is enabling agents to use APIs via a packaged agent skill (markdown + YAML name/description) and optional MCP server with a code‑execution tool. The approach uses code‑mode execution, Cloudflare workers for OpenAPI wrappers, PromptFu evals that check skill trigger and output correctness, and automated builds to keep skills current.
AI-written summary · about
chapters (17)
- 0:00Introduction to coding agents
- 1:14Evolution of API design for agents
- 3:00Rise of coding agents in development
- 4:47Two dimensions of API agent optimization
- 6:14CLI vs MCP server for API operations
- 9:00MCP server benefits and limitations
- 10:23Code execution tools for MCP servers
- 12:28Integrations with MCP and agent skills
- 13:30Agent skills for API integration
- 15:16Key elements of effective skills
- 16:38Managing skill context and files
- 18:23Skill descriptions and use cases
- 19:25Keeping skills updated and fresh
- 20:27Evaluating API skills with tests
- 21:10Example skill for Resend API
- 24:21Distributing and updating skills
- 25:43Q&A on plugin standardization
