[FEATURE] code-review plugin skips review after new commits if Claude already commented

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The new claude-code-review of GitHub Action use plugin version recently.
The code-review plugin's pre-validation step skips the review entirely if Claude has previously commented on a PR (gh pr view <PR> --comments check). This means:

1. When a PR receives an initial review and the author pushes new commits to address feedback, the synchronize event triggers the workflow but the review is skipped
2. Even if all previous review comments are resolved, no re-review occurs
3. There is no --force flag or similar option to bypass this behavior

This makes the plugin unusable for iterative PR workflows where authors push fixes and expect updated feedback. The current behavior assumes "one review per PR lifetime" which doesn't match real-world development practices.

Proposed Solution

Add a mechanism to allow re-reviews when new commits are pushed. Possible implementations:

1. Add a --force flag to bypass the "already commented" check
2. Make the skip condition smarter: only skip if no new commits exist since the last Claude comment
3. Add a --incremental flag to review only changes since the last review
4. Make the skip behavior configurable via plugin settings

The ideal solution would be option 2 (smart skip) as the default, with option 1 (--force) as an escape hatch.

Alternative Solutions

Current workarounds I've considered:

1. Revert to non-plugin approach: Write review logic directly in the prompt without using the code-review plugin, losing the multi-agent confidence scoring features
2. Delete previous Claude comments: Before each workflow run, delete Claude's old comments to bypass the check (loses review history)
3. Custom command file: Create a local .claude/commands/code-review.md that removes the skip condition (requires maintaining a fork of the plugin logic)

None of these are ideal as they either lose functionality or require significant maintenance overhead.

Priority

Medium - Would be very helpful

Feature Category

Other

Use Case Example

Typical iterative PR workflow:

1. Developer opens PR with initial implementation
2. Claude reviews and posts 3 inline comments about issues
3. Developer pushes a fix commit addressing those issues
4. GitHub Actions triggers on synchronize event
5. Expected: Claude reviews the new changes and either confirms fixes or finds new issues
6. Actual: Claude skips entirely because it already commented, developer gets no feedback

This is especially problematic for teams that rely on automated reviews as a first-pass check before human review.

Additional Context

Related issues:

• #4538 (closed as duplicate) requested "Add ability to re-run the review after fixes are pushed"
• #4539 discussed multi-commit PR review behavior
• anthropics/claude-code-action#445 - Similar problem but different root cause (reviews old diff vs skips entirely)

The skip logic is in,
https://github.com/anthropics/claude-code/blob/a6a8045031de9ff3e44683264e2ed6d434a8c0b6/plugins/code-review/commands/code-review.md?plain=1#L18

Reproduction:

• First review run (success): https://github.com/ncaq/dotfiles/actions/runs/21161212270?pr=452
• Second review run after push (skipped): https://github.com/ncaq/dotfiles/actions/runs/21166096855?pr=452