ci(claude): revert to @v1, pin --model claude-opus-4-7

[d-morrison]

Summary

• Revert the agent workflow from anthropics/claude-code-action@beta back to @v1.
• Pin --model claude-opus-4-7 in claude_args so the action stops asking the API for a retired default model.

Diagnosis (from #224's @beta debug trigger)

The @v1 SDK crashes documented in #224 were not a Claude-Code-Action bug per se — they were a model: claude-sonnet-4-20250514 404 from the Anthropic API being masked as the obfuscated sdk.mjs AJV crash. Confirmed by run 26147667637, which on @beta produced a readable error:

API Error: 404 {"type":"error","error":{"type":"not_found_error","message":"model: claude-sonnet-4-20250514"},"request_id":"req_011CbDRtTEpspPCv6cL5kaaZ"}

claude-sonnet-4-20250514 is the older Sonnet 4.0 ID that's no longer served. The action's default points there; pinning --model claude-opus-4-7 routes around it.

The upstream "obfuscated SDK crash" issue (anthropics/claude-code-action#892) is orthogonal — it's about the error message being illegible. That stays open upstream regardless of this fix.

Test plan

• Merge.
• Trigger @claude review on PR Adding population parameters as attributes #141 (or a smaller PR) and confirm a successful review run.
• If it works, drop a confirmation comment on upstream #892 noting the underlying cause for this signature.

🤖 Generated with Claude Code

[claude]

Claude finished @d-morrison's task in 50s —— View job

---

Code Review: PR #225 — ci(claude): revert to @v1, pin --model claude-opus-4-7

• Gather context (diff + workflow file)
• Understand request (code review)
• Execute review
• Post results

---

Summary: Clean, minimal fix with a well-documented root cause. The diagnosis in the PR description is solid — the retired model ID returning a 404 was being masked by the AJV error in the minified sdk.mjs, and @beta surfaced it. Good investigative work.

---

.github/workflows/claude.yml

What changed:

• Line 120: anthropics/claude-code-action@beta → @v1 (reverts temp debug change)
• Line 142: adds --model claude-opus-4-7 to claude_args; adds explanatory comment block

The fix is correct. The comment explaining the pin and its cleanup condition is well-written and gives future maintainers exactly what they need.

---

Observations

1. Model tier choice (minor)

claude-opus-4-7 is the highest-capability (and highest-cost) model tier. For routine CI code review and issue-filing tasks, claude-sonnet-4-6 would be sufficient and meaningfully cheaper per token. Consider whether Opus is warranted here vs. Sonnet.

Fix this →

2. Comment placement (cosmetic)

The new comment block (lines 136–141) is inserted between the existing Bash-allowlist comment and the claude_args value itself. This is readable, but the comment now describes two separate concerns (Bash tools AND the model pin). No action needed — just noting it reads fine.

3. Removed temp comment (good)

The four-line # TEMP: swapped @v1 → @beta... block is cleanly removed. No stale documentation left behind.

---

Verdict

Approved. The change is correct, minimal, and well-commented. The only non-blocking suggestion is to evaluate claude-sonnet-4-6 vs. claude-opus-4-7 for cost efficiency once the fix is confirmed working.

---

[codecov]

Codecov Report

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