BLOG / SETUP, SECURITY AND AUTOMATION

Share Project Instructions Across AI Coding Agents

One repository guide connected to several AI coding tools
One repository guide connected to several AI coding tools

A team may use several coding agents in the same repository. One developer opens an editor assistant, another works from a terminal CLI, and a third asks a cloud agent to prepare a pull request. They need the same project decisions, but the tools may look for different instruction files or apply different rules.

Separate shared project knowledge from tool specific setup

Start with the facts every tool should know: how the project is structured, which commands validate a change, where generated files live, and which compatibility constraints matter. Keep this guidance in a reviewed source file. Then add a small adapter or link for each tool that needs a different filename or location.

Do not copy a long rule set into several files and expect every copy to stay current. If a tool supports an include or reference mechanism, use its documented syntax. If it does not, keep the tool specific file short and add a check that prompts someone to update it when the shared guide changes.

Instruction file names depend on the agent and its mode.
Tool or environmentDocumented project guidance examplesWhat to verify
OpenAI CodexAGENTS.mdWhether nested files and the current CLI version are supported
Anthropic Claude CodeCLAUDE.md and supported rulesHow the installed version discovers and imports files
GitHub Copilot.github/copilot-instructions.md and supported agent filesWhether the selected feature reads the file
Visual Studio CodeAGENTS.md, CLAUDE.md, or Copilot instruction files by harnessWhich harness and instruction setting is active

Test that each agent actually sees the guidance

Use a harmless question that has a specific answer in the project instructions. Ask each tool to identify the relevant file and summarize the rule it will follow. This checks discovery; it does not prove the model will obey every instruction on every task.

Put mandatory controls in code, permissions, or CI. An instruction file can explain that a production command needs approval, but it cannot prevent a tool from running that command if the environment gives it permission.

Keep the shared guidance short and maintained

Prefer durable project decisions over generic advice that applies to any repository. Add a rule when it prevents a real mistake, and remove it when the code or workflow changes. Keep test commands accurate and note where a developer must ask before taking an action.

LucentraCode documents project configuration and instruction support in its current reference. Check that documentation for the installed version, and use the same source of truth your team maintains for its other coding agents.

Sources and further reading