Config file

When you will meet it

The first time you configure an AI tool you are almost always editing a config file, not clicking a GUI: settings.json, a .env for keys, an MCP server registry. Guides say "paste this into the config" — without knowing where that file lives, what format it uses and how to recover a broken edit, you are stuck at that one sentence. Reading config files is also the "write the setting once, keep it forever" skill that frees you from retyping flags.

Minimal example

# 每個設定都是從某個檔案讀來的 —— 這個指令把來源標在每一條前面：
git config --list --show-origin

Every output line starts with the file that setting was read from — changing configuration literally means editing those files. Nearly every program has such a location; its documentation tells you where.

What people get wrong

• Editing config while the program runs and wondering why nothing changed. Many programs read config only at startup — and some write the file back on exit, overwriting your edit. Restart after changing, as the docs say.
• Skipping format validation after editing JSON/YAML configs. One missing comma or stray tab, and the failure is often just "failed to load config" with no line number. Run JSON through python3 -m json.tool; for YAML, check indentation first.