Skip to content

jszmajda/how_i_claude

Repository files navigation

How to Work with Agentic AI Coding Assistants

Practical guides for getting effective results with Claude Code, Cursor, Windsurf, and other agentic AI coding tools.

Based on analysis of 1,352 real development messages to identify what communication patterns lead to productive collaboration.

Guides

Skills

The skills/ folder contains reusable skill definitions for Claude Code:

  • linked-intent-dev - Linked-intent development (LID) — structured workflow for design-before-code: HLD → LLD → EARS specs → Implementation plan

Key Insights

The shift from Copilot to agentic AI:

  • Copilot: You write code, AI suggests completions
  • Agentic AI: AI writes code, you steer and validate

What makes sessions effective:

  1. Be detailed - "I see X when I expected Y" not "it's broken"
  2. Close the loop - Report results back, even just "yup"
  3. Steer constantly - "Great. Now let's..." every few exchanges
  4. Validate assumptions - "What did you understand?" after work is done

The real anti-pattern isn't "too many turns" - it's frustrated circling. Exploration with back-and-forth is fine. Repeating "try again" without new information is the problem.

Data

The analysis_data_index.md file describes the raw data files used in this analysis, with query examples.

License

MIT - See LICENSE

About

Notes and skills for how I use Claude Code

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors