Claude PR reviews are influenced by other comments

We're using claude-code-action to do automatic PR reviews and can see that generated reviews are influenced by other comments, including comments from claude itself. For example

if someone left a comment saying that certain line won't work, claude will just assume that this is a valid comment without actually verifying this statement and claiming that this is a "well founded feedback that need to be addressed" causing confusion. not to mention, that comments at times may be excessive and off topic.

I _think_ the reason this is happening is that claude gets all PR details as prompt - [link](https://github.com/anthropics/claude-code-action/blob/main/src/create-prompt/index.ts#L513)

Attempts to add something along the lines `**Ignore all prior comments, discussions, or reviews**, even from yourself.` to the direct_prompt didn't work.

Im wondering if there is a way to work around that and make claude reviews truly independent?

Or, for example, i can look into making a PR to allow PR context customisation:

```
- name: review
uses: anthropics/claude-code-action@v0.0.32
with:
anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
timeout_minutes: '15'
use_sticky_comment: false
inject_context:
- pr_title
- pr_description
- comments
- review_comments
- ...
direct_prompt: Review this PR
```

Or, in fact, any other implementation we can agree on to solve this problem.