How is create-release-branch different from release?

I would suggest changing the options to

• rc
• patch
• minor

These are usually the three release types we do.

Patch and minor are the same steps to be taken here. RC is from main, minor/patch happen on release branches, and the 3rd action release managers have to take is creating the branch in the first place. You would run create-release-branch at the beginning of code freeze, and then run release at the end for a minor.

Ideally, the release freeze should be time-based and automatic.

Looks like gitRun tosses all output from the git command. Did you mean to direct the output somewhere?

P1 [CRF-1] Echoing the broader concern here: scripts/release-action/ has 1339 lines of new Go code with 96 lines of tests (14:1 ratio). Core release logic is untested.

The only tests cover parseVersion, version.String(), and version.IsRC(). At least 16 pure functions that need no git state to test have zero coverage: versionIsLess, findLatestRC, findLatestNonRC, findPreviousTag, filterTagsForSeries, isStable, isHexSHA, humanizeTitle, categorizeCommit, parsePRNumbers, stripPRRef, isDependabot, commitSortPrefix, ReleaseRequest.String, CreateBranchRequest.String.

These implement the version calculation and release note generation logic that determines what gets tagged and published. A bug in versionIsLess or findPreviousTag produces wrong tags in production. (Netero)

🤖

Added unit tests for all 15 pure functions: versionIsLess, findLatestRC, findLatestNonRC, findPreviousTag, filterTagsForSeries, isStable, isHexSHA, humanizeTitle, categorizeCommit, commitSortPrefix, parsePRNumbers, stripPRRef, isDependabot across calculate_test.go and commit_test.go.

Generated with Coder Agents

Good catch. Fixed the comment and removed the redundant nil assignments (they're already the zero value). gitRun intentionally discards output; it's only used for exit-code checks like merge-base --is-ancestor.

Generated with Coder Agents

Seems like this is an object. Quick test on my machine:

$ gh pr list --json "number,title,author,url" --limit 1
[
  {
    "author": {
      "id": "U_kgDODaQKsg",
      "is_bot": false,
      "login": "tracyjohnsonux",
      "name": "TJ"
    },
    "number": 25595,
    "title": "refactor(site): promote search to full-width sidebar nav item",
    "url": "https://github.com/coder/coder/pull/25595"
  }
]

Bug confirmed, thanks for the repro. Fixed to parse author as struct{ Login string } matching gh's actual output.

Generated with Coder Agents

Planning to remove these old scripts in a follow-up PR?

Keeping the old scripts in place as a fallback. No deletions in this PR.

Generated with Coder Agents

P2 [CRF-2] Author is declared as string but gh pr list --json author returns a JSON object {"login": "username"}. Go's json.Unmarshal silently skips type-mismatched fields, so pr.Author is always "". The error message at line 109 renders (by @) instead of (by @username).

Compare with ghBuildPullRequestMap at line 54 in this same file, which correctly uses Author struct { Login string }. Fix: change line 95 to match the struct pattern at line 54 and reference .Author.Login at line 109.

🤖

P2 [CRF-3] "To test the workflow without publishing, select dry-run" references a removed feature. The dry_run input was deleted from the workflow in this PR. Either remove this line or describe the current way to test (if one exists).

🤖

Nit [CRF-4] regexp.MustCompile inside function body. Every other regex in this package (branchRe, cherryPickPRRe, conventionalPrefixRe, breakingCommitRe, prNumRe, semverRe) is a package-level var compiled once at init. This one recompiles on every call to categorizeCommit. Not a performance concern for a release tool, but it breaks the file's own convention. Move it to a var block.

🤖