ci: upgrade pr-reviewer-action to v1.2.3 (tool loop, carry-forward verdicts)

[joryirving]

Upgrades to pr-reviewer-action v1.2.3.

What this brings

• Cumulative incremental verdicts: when a review requests changes, its findings are carried into the next push's incremental review, which must explicitly resolve each one — fixing one of three blockers can no longer rubber-stamp the other two. The latest review always lists what was resolved and what's still open.
• Superseded reviews keep their content: old reviews are hidden (minimized) and their verdicts dismissed, but the text is no longer overwritten — expanding a hidden review still shows what it said.
• Multi-turn tool research — tool_mode: plan_execute_loop with tool_max_rounds: "2": the model can now react to what its evidence tools returned and request follow-ups (read the changed file → grep for its callers). The 4-request budget is now shared across rounds; each extra round costs one additional planning call against the review endpoint.
• Explicit smart wiring: ai_smart_model/ai_smart_api_format now read vars.SMART_MODEL/vars.SMART_FORMAT directly (both are explicitly set at the org level), instead of coalescing to the fallback vars.
• Dirty-baseline escalation (on by default): incremental reviews of a PR whose previous review found issues run on the smart model.

🤖 Generated with Claude Code

[its-saffron]

AI Automated Review

Full PR review.

Analysis engine: review@https://litellm.jory.dev/v1 (openai)

Recommendation

Approved. This PR upgrades the pr-reviewer-action to version v1.2.3, enabling advanced agentic capabilities like multi-turn tool loops and cumulative incremental verdicts.

Change-by-Change Findings

• .github/workflows/ai-pr-review.yaml
  • Upgraded misospace/pr-reviewer-action from v1.2.2 to v1.2.3 (using specific SHA 59c7a0596b0d0b515ef9137f7acb985115f4c1d8).
  • Updated ai_smart_api_format and ai_smart_model to use direct environment variables (vars.SMART_FORMAT and vars.SMART_MODEL) instead of falling back to defaults. This aligns with the release notes indicating explicit smart wiring.
  • Enabled tool_mode: plan_execute_loop and set tool_max_rounds: "2". This enables the model to react to tool results and request follow-ups, as described in the v1.2.3 release notes.

Standards Compliance

• Security: No secrets are committed. The workflow uses existing GITHUB_TOKEN and LITELLM_API_KEY secrets.
• Conventions: The change follows the repository's pattern of using specific SHAs for workflow actions to ensure deterministic builds/runs.

Linked Issue Fit

• No linked issues were provided in the PR description.

Unknowns or Needs Verification

• None. The changes are a straightforward dependency upgrade and configuration update to support new features in the pr-reviewer-action.