ci(claude): temporarily swap @v1 → @beta to surface real SDK error

[d-morrison]

Summary

• Swap the agent workflow's anthropics/claude-code-action@v1 → @beta for one debug run.
• Goal: surface the real underlying error currently being masked as an obfuscated minified-JS frame from sdk.mjs (matches upstream issue #892).

Motivation

The @claude review workflow has been crashing during SDK init with no readable error — three deterministic failures (runs 26144000481, 26144129682, 26146565378) on PR #141, all at the same byte offset (sdk.mjs:60:11252) in the SDK's auth/credential init.

Adding show_full_output: true did not help, because the SDK throws before the child writes any stderr that could be piped through.

A maintainer-thread comment on #892 (ovceev, 2026-03-10) notes that @beta does surface the underlying error (in their case, Credit balance is too low). One debug trigger on @beta should tell us what's actually killing the child.

Test plan

• Merge this PR.
• Trigger @claude review on PR Adding population parameters as attributes #141 once.
• Capture the real error from the action log.
• Open a follow-up PR to revert to @v1 (and decide whether to file or update an upstream issue based on what the error reveals).

🤖 Generated with Claude Code

[claude]

Claude finished @d-morrison's task in 1m 35s —— View job

---

Code Review: PR #224 — @beta debug swap

• Gather context
• Read changed file
• Review diff
• Post findings

---

Summary

This is a minimal, well-motivated diagnostic change — a single-line action version swap with a clear revert plan. The diff is clean and the inline comment is exemplary for a temporary change.

What changed

.github/workflows/claude.yml:124

-        uses: anthropics/claude-code-action@v1
+        uses: anthropics/claude-code-action@beta

Plus a 4-line explanatory comment with upstream issue reference and revert condition.

Strengths

• Comment quality: The inline comment explains why (@v1's AJV minification masks the error), cites the upstream issue (#892) and the specific commenter/date, and states the revert condition. This is the right pattern for temporary changes.
• Minimal diff: Only the action tag changes. All other settings (show_full_output: true, claude_args, permissions) are preserved — maximizing the chance of capturing the real error.
• Explicit test plan: The PR body lays out a clear 4-step plan including the revert PR. Well done.

One concern worth flagging

@beta affects every @claude trigger on main, not just PR #141.

The workflow fires on any @claude mention across all issues and PRs. Once merged, any team member who triggers Claude during the debug window gets @beta behavior. This is probably fine for a short window, but if the revert is delayed (e.g., a weekend), multiple unrelated runs will hit @beta.

Suggestion: File a GitHub issue immediately after merge (before triggering the debug run) as a revert reminder — something like chore: revert claude-code-action to @v1 after #224 debug. That way the revert tracking doesn't depend solely on the PR test plan.

Minor note

Both @v1 and @beta are mutable floating tags — neither is SHA-pinned. So the risk delta of this swap is small; this isn't introducing a new category of mutability. If the repo ever adopts SHA pinning for actions (a good practice), both would need updating.

Verdict

Looks good to merge. The change is safe, well-documented, and the approach (using @beta to surface masked errors) is validated by the upstream issue thread. No code logic is affected — this is purely a version pointer swap on a diagnostic path with a clear revert plan.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.