Problem
action.yml pins a specific modelith release as its version input default
(see CLAUDE.md's "Cutting a release" checklist). That bump is a manual step
after every release — easy to forget, and nothing currently catches it if it
is. A forgotten bump doesn't error; it just silently leaves the action
installing an old release indefinitely.
Proposed fix
Add a CI check (or a step in release.yml itself) that fails if
action.yml's version default doesn't match the most recent release tag.
Options to weigh:
- A check in
ci.yml that runs on every PR touching action.yml or on a
schedule, comparing the pinned default against gh release view --json tagName (or the GitHub API equivalent).
- A step at the end of
release.yml that bumps action.yml automatically and
opens a PR (or commits directly, mirroring how the Homebrew formula is
pushed automatically today).
The second option removes the manual step entirely rather than just catching
it after the fact — probably the better end state, but more work and a
different failure mode (auto-commit vs. CI gate) to think through.
Context / priority
Not urgent — there's currently exactly one release (v0.4.0) and the default
is correct. Worth doing before this becomes a recurring manual step that
eventually gets missed.
Problem
action.ymlpins a specificmodelithrelease as itsversioninput default(see
CLAUDE.md's "Cutting a release" checklist). That bump is a manual stepafter every release — easy to forget, and nothing currently catches it if it
is. A forgotten bump doesn't error; it just silently leaves the action
installing an old release indefinitely.
Proposed fix
Add a CI check (or a step in
release.ymlitself) that fails ifaction.yml'sversiondefault doesn't match the most recent release tag.Options to weigh:
ci.ymlthat runs on every PR touchingaction.ymlor on aschedule, comparing the pinned default against
gh release view --json tagName(or the GitHub API equivalent).release.ymlthat bumpsaction.ymlautomatically andopens a PR (or commits directly, mirroring how the Homebrew formula is
pushed automatically today).
The second option removes the manual step entirely rather than just catching
it after the fact — probably the better end state, but more work and a
different failure mode (auto-commit vs. CI gate) to think through.
Context / priority
Not urgent — there's currently exactly one release (
v0.4.0) and the defaultis correct. Worth doing before this becomes a recurring manual step that
eventually gets missed.