chore: update auto reviewer

[finnar-bin]

update auto reviewer

[github-actions]

Code Review: chore/update-auto-reviewer

Overview

This PR makes three targeted improvements to the Claude auto-reviewer workflow:

1. Enables sticky comments to avoid duplicate review comments on each push
2. Pins the Claude model to claude-sonnet-4-6
3. Increases max turns from the default to 20

---

What's Good

• use_sticky_comment: true is a clear improvement. The workflow triggers on synchronize events, so without this, every push generates a new top-level review comment — sticky ensures the previous one gets updated instead, keeping PR threads clean.
• The existing timeout-minutes: 15 and cancel-in-progress: true concurrency guard already limit runaway cost/time, which provides a reasonable safety net alongside --max-turns 20.

---

Concerns

Model pinning — Pinning --model claude-sonnet-4-6 by name without a dated version suffix (e.g., claude-sonnet-4-6-20250219) means the resolved model can silently shift if Anthropic reassigns the alias. More practically, when this model is retired the workflow will fail with an opaque API error. Worth adding a comment in the YAML or a tracking note to revisit the model periodically.

--max-turns 20 — Higher than a typical default (4). On a busy repo this workflow fires on every commit push, so a large PR with many files could rack up 20 tool-call turns per sync event. Consider whether a lower ceiling (e.g. 10–12) would still produce thorough reviews while keeping API costs more predictable.

---

Verdict

Changes are safe and well-scoped. The sticky-comment fix alone is worth the merge. The two concerns above are worth a quick discussion before landing, but neither is a blocker.

[github-actions]

Code Review

Overview
This PR makes three small improvements to the Claude auto-reviewer workflow: enabling sticky comments, pinning an explicit model, and raising the max-turns limit.

---

Positives

• use_sticky_comment: true — Good UX improvement. Instead of appending a new top-level comment on every push, the action will update the existing review comment in place, keeping PR threads clean. Works cleanly with the existing cancel-in-progress: true concurrency setting.
• --max-turns 20 — Reasonable increase from the default (10). Gives the model more headroom to complete reviews on larger PRs.
• The action is already pinned to a full commit SHA (1dc994ee...), which is a security best practice and is unchanged.

---

Considerations

• Model pinning (--model claude-sonnet-4-6): Explicit pinning ensures reproducible behavior, but it means the workflow won't automatically benefit from newer model releases. The team should revisit this when a newer Sonnet version ships. Consider adding a comment in the file noting when this was pinned, or using a versionless alias if the action supports one.

• Cost impact of --max-turns 20: Doubling the turn budget doubles the worst-case API cost per review. For typical PR sizes this is likely fine, but worth monitoring on large PRs.

• claude_args as a multiline YAML string: The existing format (one CLI flag per line) is already in use and works, so the additions are consistent. No issues here.

---

Summary

Small, clean change. No bugs or security issues. The sticky-comment improvement is the most valuable part. The model pin is a minor maintenance burden to accept. Looks good to merge.