Codex

Codex reads AGENTS.md from the project, and cordango new writes one with the same content as CLAUDE.md. There is nothing to configure.

Create the workspace

mkdir acme && cd acme
cordango new support

Start Codex in it

codex

Sandbox settings

A Cordango workspace is a good fit for workspace-write. The whole authoring loop writes only inside the workspace, and the parts that reach the network are the four connected commands, which you will want to approve anyway.

# .codex/config.toml
sandbox_mode = "workspace-write"
approval_policy = "on-request"

That leaves cordango publish and cordango login asking before they touch an instance, which is the right default for a command that changes what a real workspace is running.

Starting prompt

Add a "priority" field to the ticket entity: a select with low, normal, high and urgent,
defaulting to normal. Show it as a column on the tickets table.

Use cordango vocabulary field if you need to check what a select accepts. Run cordango check when
you are done and show me git diff. Do not commit.

Starting from nothing

Build a support desk app in this workspace.

The job: a small team receives customer problems, works them, and needs to see what is overdue.
Tickets belong to a customer company, have an owner, move through new / in progress / waiting on
customer / resolved, and carry a priority.

Steps, in order:
1. Run cordango inspect and tell me which core apps exist. Reference them rather than modelling
   your own companies or people.
2. Show me the entity list you intend to create, and wait for me to agree.
3. Create them one aggregate at a time, running cordango check after each.
4. Add the lifecycle, then the roles, then the screens.
5. Finish with cordango check and git diff.

Do not commit.

Non-interactive runs

codex exec runs a prompt without the TUI, which makes the check loop scriptable:

codex exec "Run cordango check --json and summarise any errors by file. Do not change anything."

Keep an exec prompt read-only unless you have a reason not to. An agent editing files with nobody watching is a fine way to discover that the aggregate you scoped it to was the wrong one.

Nested instructions

Codex reads AGENTS.md from subdirectories as well as the root. That is useful in a workspace with several apps: a per-app apps/support/AGENTS.md can carry the business rules that only apply to that app, while the generated root file keeps carrying the Cordango workflow. Do not restate the workflow in a nested file. Add to it.

Get started

Quickstart with an AI agent

Concepts

Guides

Security

Sandbox settings

Starting prompt

Starting from nothing

Non-interactive runs

Nested instructions

​Sandbox settings

​Starting prompt

​Starting from nothing

​Non-interactive runs

​Nested instructions

Sandbox settings

Starting prompt

Starting from nothing

Non-interactive runs

Nested instructions