Your own rules for AI agents

Cursor, Claude Code, Codex and Antigravity read the same instruction files from local-configuration; if you change them there and push, they apply to the whole team as soon as the setup app or ap platform projects clone has updated the project.

The instruction files

Your own rules belong in AGENTS.md; change the tool-specific files only for rules that concern a single tool.

What is already in there

The generated AGENTS.md contains the platform ground rules: start via ap run-local, no changes to .gitlab-ci.yml and env/*.generated.env, custom values in env/*.custom.env, no secrets in documentation, logs, commits or chat replies, and work in the matching app repository, not in local-configuration. Leave these sections in place.

Adding your own rules

Describe your way of working in its own section of local-configuration/AGENTS.md, for example so that the agent commits and pushes on its own:

## Git workflow

- Create a commit after every completed task and push it.
- Use Conventional Commits (feat, fix, chore, docs, refactor, test).
- Work on feature branches and never commit directly to main.

Commit and push the change in local-configuration; on the next clone or update it applies to the whole team. By default, agents do not commit on their own, and the restriction to feature branches prevents an accidental deployment from a push to main.

Helpers around Git

ap git auto-commit stages, commits and pushes in one step and works as an instruction in a rule. The pre-commit hook ap git pre-commit check from the setup app blocks commits to protected files, even if an agent changed them. What happens after a push to main is covered in Git workflow and deployment.