Shared VS Code Settings for AI-Assisted Teams

A shared coding assistant workflow needs more than a preferred model. Start with a workspace contract: what the editor standardizes, what the assistant should follow, and what stays private. This Quanta Briefings walkthrough builds that contract without turning personal editor preferences into team policy.

01 · Step 1: Put repository settings under version control

Open the project folder, then run Preferences: Open Workspace Settings (JSON) from the Command Palette. For a single-folder workspace, VS Code stores these settings in .vscode/settings.json. A saved multi-root workspace can also hold settings in its .code-workspace file. Choose the structure your team actually opens, rather than maintaining competing configurations.

Commit only settings that affect collaboration. Keep themes, font sizes, and personal shortcuts outside the repository. Review workspace changes through pull requests, with an explanation of which recurring problem each setting solves.

02 · Step 2: Standardize formatting before assistant behavior

Use the Workspace Settings editor to configure editor.formatOnSave, editor.insertSpaces, and editor.tabSize according to the repository’s existing conventions. Select the agreed formatter through editor.defaultFormatter; use language-specific settings when different languages need different tools. VS Code supports workspace and language-specific configuration rather than requiring one global formatting choice.

• › Match the project’s formatter configuration instead of introducing a second style policy.
• › Test a small assistant-generated edit, save it, and inspect the diff for unrelated changes.
• › Keep formatting and lint checks in continuous integration as the acceptance gate.

03 · Step 3: Give the assistant explicit repository instructions

Separate editor configuration from coding guidance. For GitHub Copilot in VS Code, place project-wide guidance in .github/copilot-instructions.md. Other assistants may discover different files, so verify the instruction mechanism for each tool your team supports rather than assuming one file works everywhere.

Write short, actionable rules: where business logic belongs, which test command to run, how to handle generated files, and when dependency changes require approval. Ask the assistant to identify assumptions and report checks it could not execute. Treat these instructions as guidance to validate, not permission controls.

04 · Step 4: Narrow searches without confusing exclusions with security

Add generated output and bulky directories to search.exclude where appropriate. Use files.exclude deliberately when files should also disappear from the Explorer. VS Code documents these settings for controlling search and visibility, and its integrated assistant has documented workspace-index exclusions.

Do not treat those patterns as a universal assistant firewall. Test your actual extension or terminal-based tool, including direct file access. Keep credentials outside tracked files, review the context supplied to requests, and establish which repositories and data categories may leave the development environment.

05 · Step 5: Separate shared routing from private credentials

For sustained workloads across supported clients, consider Quanta as the shared API relay. It provides one pay-as-you-go balance across models and works with Claude Code, Codex, Gemini CLI, and OpenAI SDK integrations. Open a Quanta account, then document the API base address, https://api.quantarelay.com, in your team’s setup guide.

Configure that address through each client’s supported mechanism; do not invent a universal VS Code setting for API routing. Keep API keys in the client’s supported private credential mechanism, never committed workspace settings. Record the chosen model identifier separately so routing changes remain reviewable.

Quanta meters calls on real tokens, with cache writes and cache reads priced separately. Include those categories when reviewing usage instead of judging cost by request count alone.

06 · Step 6: Validate the workspace from a fresh checkout

Ask a teammate to follow the setup guide without copying your personal settings. Keep Workspace Trust enabled and inspect unfamiliar repositories before trusting them; Restricted Mode limits features that could execute project code.

Finally, run one bounded assistant task: change a function, add a test, save, and execute the documented checks. Inspect formatting, instruction adherence, credential handling, and the final diff. Commit the configuration only after this rehearsal, and assign an owner to revisit it when tools or repository conventions change.