v1.0.3

Highlights

• Hardened inline review prompt to significantly reduce invalid pending review submissions.
• Enforced canonical payload rules to prevent missing required fields (e.g., pullNumber, line ranges).

Improvements

• Refactored inline comment policy to make range-line comments the default and strictly limit single-line exceptions.
• Clarified review body handling to use plain Markdown with real newlines, avoiding escaped text issues.
• Increased max-turn limits to improve review completeness on deeper diffs.

Fixes

• Fixed an issue where inline review bodies were treated as JSON strings, causing malformed comments.
• Improved robustness of pending review creation by preventing tool calls when required fields are missing.

CI / Action Changes

• Updated the “Run Claude Code review” step to continue on error to avoid blocking workflows.
• Added automatic upload of Claude logs for easier debugging and traceability.

Maintenance

• Removed unused files and performed minor prompt formatting cleanups.

v1.0.2

Fixed

• Prevent markdown code blocks in prompt scripts from being executed as shell commands.
• Stabilize inline comment prompt generation after v1.0.1.

v1.0.1

Fixed

• Fix bash heredoc parsing error caused by CRLF line endings in GitHub Actions environment.

v1.0.0

🎉 Initial Release (v1)

✨ Features

• Simple LLM-based code review without additional setup
• Automatically generate review prompts from PRs, linked issues, and reference files
• Post code review comments using Claude Code
• Support multiple review languages via configuration

🚀 How to Use

• Add @claude to the pull request body or comments to trigger the review

📦 Notes

• First public release on GitHub Marketplace