Skills vs. Subagents in Cursor: Build Smarter AI Workflows (Not Just Better Prompts) скачать в хорошем качестве

Skills vs. Subagents in Cursor: Build Smarter AI Workflows (Not Just Better Prompts)

Cursor isn't just an AI vibe code editor — it's a platform for building repeatable, composable workflows using Skills and Subagents. This talk breaks down the difference and shows you how to build both from scratch. What we cover: 🔷 Skills — markdown files Cursor dynamically reads before taking action. They encode HOW to do a task (steps, constraints, output format). 🔶 Subagents — autonomous markdown agents with a defined trigger, input, output, and isolation boundary. They encode WHO does a task and WHEN. 🎬 Demo A — using a verifier subagent in Cursor 🎬 Demo B — using a git-commit skill We also go deeper: → Level 2 multi-model workflows: Opus as architect, cheaper models as coders and verifiers → The Planner / Subplanner / Worker / Git hierarchy (from Cursor's own self-driving codebases research) → Why a generic "programmer agent" is an antipattern — and the 5-question litmus test (TRIGGER / INPUT / CONSTRAINT / OUTPUT / ISOLATION) to know if your worker agent is legit By the end you'll know exactly what to build first and why. 📁 Resources: 📖 Cursor Docs: https://cursor.com/docs/skills https://cursor.com/docs/subagents 📖 Cursor blog: https://cursor.com/blog/self-driving-codeb...