Recommend strict `files.exclude` and `search.exclude` to minimize noise in `#textSearch` results and reduce token consumption

[tamuratak]

Since Copilot has transitioned to usage-based billing, it is recommended to document best practices for controlling token consumption somewhere.

To prevent the built-in tool grep_search, or #textSearch, from returning irrelevant results and filling the context window with noise, it is advisable to strictly configure files.exclude and search.exclude. This is recommended because grep_search respects these settings by default, which can help keep junk out of the search results.

Even if the agent does not read a file in the search results, if the file is in a format like jsonl, the search results alone will consume many tokens.

---

See:

• https://github.com/microsoft/vscode/blob/fdabaa87e945796e7abb30e92ad3d31c94288436/extensions/copilot/package.json#L395

"Whether to include files that would normally be ignored according to .gitignore, other ignore files and files.exclude and search.exclude settings. Warning: using this may cause the search to be slower. Only set it when you want to search in ignored folders like node_modules or build outputs."