[brentvatne]

Why

templates/dismiss.yml defaulted to ECR_VERSION: ${{ vars.ECR_VERSION || '^0.2.3' }}. On 0.x, ^0.2.3 means >=0.2.3 <0.3.0, so every repo scaffolded by ecr init without an ECR_VERSION variable runs /dismiss on 0.2.3, 33 releases back. The other three templates default to 'latest'.

What that does today:

• 0.2.3 writes back only { review, dismissed }, so each /dismiss silently drops feedback, pins, scopes and inline links from the comment state.
• 0.2.3's renderer throws on a severity it doesn't know, so once feat: add design and nit severities #86's design and nit land, /dismiss fails outright.

expo/expo isn't affected (it pins through its own script), and expo/terraform-iam pins an exact version.

What

• Default to 'latest', with a comment on why /dismiss has to track the review workflows.
• A test in init-scaffold.test.ts that every templates/*.yml defaults ECR_VERSION to latest. It fails with the old default.

Existing adopters keep their copy of the workflow, so they need to update it, or set the ECR_VERSION variable. Worth a line in the release notes.

🤖 Generated with Claude Code

[github-actions]

🤖 AI code review

Decision: Ready for human review

Overall PR risk: Low. The change modifies the /dismiss workflow default from ^0.2.3 to latest, aligning it with the other three workflow templates (atlantis.yml, command.yml, workflow.yml). Only new scaffolds and re-scaffolds are affected; existing adopters keep their current workflow and can update manually or pin versions via the ECR_VERSION variable. The main risk is that engine regressions now reach /dismiss immediately, but this is mitigated by available pinning mechanisms.

🔗 Review setup:

• This PR changes code the reviewer prompts cite (src, src/tests, templates). Check that the guidance quoting it is still correct.

No findings.

---

This review is advisory — it never blocks a merge and never auto-approves.