Project instructions
Project instructions define "how we work here": conventions, constraints, preferred commands, and review expectations. They help agents produce consistent results across tasks and reduce the need to repeat the same guidance in every prompt.
What Air uses as project instructions
Agents pick up instruction files and configuration that already exist in your project. For example:
CLAUDE.mdand the.claudefolder for Claude Agent, including settings files such as.claude/settings.json.AGENTS.mdfor agent workflows and reusable guidance.Other agent-specific files that are supported by the selected agent.
There is no separate instruction format to learn. If your project is prepared for agentic development, the selected agent gets those files as they are.
This means you can keep instructions in version control, share them with the team, and evolve them alongside the code.
Because these are repository files, they apply wherever the task runs. A cloud task gets them with the checkout, like any other file in the project.
Project-specific AI settings
These settings are part of the Air app. The browser has no settings screen inside a task, so there is nothing to set per project there – see What the browser doesn't have.
In Settings, the tab named after your project holds settings that apply only to the current workspace. Its section keeps agent configuration with the project:
Agent review prompt – instructions that guide how the agent reviews your code. See Review with agent.
Setup script under Docker and Worktree – instructions for initializing a run environment. See Run environments.
Open project AI settings
Open Settings.
Switch to the tab with your project name.
Go to the section.
Both are stored in the .air directory in your project root, so you can commit them and share them with the team – see Share workspace settings through version control.