Claude auto reviewer: set model and sticky comment

[mattezra]

No description provided.

[github-actions]

Code Review

Overview

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

1. Enables sticky comments so repeated pushes update a single comment rather than flooding the PR thread.
2. Explicitly pins the Claude model to claude-sonnet-4-6.
3. Caps agent turns at 25 to bound API costs.

Assessment

Sticky comments (use_sticky_comment: true) — Good change. Without this, every synchronize event (new push) creates a fresh review comment, which clutters the PR thread. Updating in-place is cleaner.

Model pinning (--model claude-sonnet-4-6) — Reasonable for a CI workflow where reproducibility matters. Trade-off is that it won't automatically pick up newer/better models. Consider a periodic review to bump this when a meaningfully better model is available.

--max-turns 25 — A sensible safety rail. The current allowedTools list is narrow (view/diff/comment only), so a complex review is unlikely to need anywhere near 25 turns in practice. No concern here.

Minor Notes

• The claude_args block passes flags as a multi-line string; the action presumably joins these lines. Worth confirming this is the documented input format for the action version pinned here (v1.0.127), but it appears to already work given the existing --allowedTools line uses the same style.
• No secrets, credentials, or sensitive data are touched by this change.

Verdict

Clean, low-risk configuration improvement. No blocking issues.