Skip to main content

Changesets

Some commands don't just report a result — they propose changes to your files. A code generator produces new source, an autofixing linter rewrites formatting. Dagger returns these as a changeset: a diff you can review before anything touches your working tree.

Changesets come from generators (run by dagger generate) and from functions written to return one — a formatter's fix, for example, called with dagger api call. A check never produces a changeset; it only validates. dagger check can verify that a generator's output is already up to date, but it never applies a fix itself.

Review and apply​

When a command returns a changeset, Dagger shows you a summary of the changed paths and line counts, then waits for you to apply it. Nothing is written until you approve.

Apply automatically​

Pass -y / --auto-apply to apply changes without prompting — useful in scripts and non-interactive sessions:

dagger generate -y

Coding agents and non-interactive use​

When Dagger detects that a coding agent is running dagger generate, it requires an explicit choice before starting any generators. Pass -y / --auto-apply to apply the result, or pass --no-apply to run the generators and show the changes without writing them to your working tree:

dagger generate --no-apply

--no-apply exits successfully even when it finds pending changes, just like choosing Discard at the interactive prompt. It only prevents Dagger from exporting the returned changeset; generators still run and may perform other work. To make pending generated changes fail the command, use dagger check --generate.

In CI​

In CI you typically want to verify that committed files are already up to date rather than rewrite them. dagger check runs your checks and also runs each generator as a read-only check — it fails, without applying anything, if a generator's output differs from what's committed:

# GitHub Actions
- run: dagger check

Use dagger check --generate to run only the generator verification. A failing generator check means the committed output was stale — run dagger generate (or the relevant fix function) locally to apply the changeset, then commit.