Best Practices for AI Coding Assistants

Chapter 2 · Practice 1 of 4

Name the files, the scenario and what "fixed" looks like

From Writing task prompts

Why it matters

Claude Code's docs show the difference specific context makes:

VagueSpecific
add tests for foo.pywrite a test for foo.py covering the case where the user is logged out; avoid mocks
fix the login bugusers report login fails after session timeout; check the token refresh in src/auth/; write a failing test that reproduces it, then fix it
add a calendar widgetlook at how existing widgets on the home page are built and follow that pattern

The specific versions do four things: scope the task, point to sources, reference existing patterns and describe the symptom with what done looks like.

GitHub's Copilot guide gives the same advice: be specific about requirements and provide examples of inputs, outputs and implementations.

Do

  • Point to an existing file that shows the pattern to follow

Avoid

  • "Fix the bug" with no symptom, location or definition of fixed