Plannotator drops custom reviews, new engines

Allowing developers to map arbitrary Agent Skills to code review tasks turns Plannotator from a simple visual diff viewer into a customizable, automated quality gate. By incorporating Cursor and OpenCode via a unified protocol, it consolidates disparate AI dev tools under a single interface.

• –**Custom Skill Reviews**: Users can now target specific security, formatting, or architectural guidelines by running standard agent skills directly against code changes.
• –**Unified Engine Marker Protocol**: Cursor (agent) and OpenCode (opencode run) CLIs are parsed via a delimited findings format, offering a consistent experience across all supported LLM backends.
• –**Improved Annotation Persistence**: Deleted reviews now use a generation tombstone to ensure deletions survive a page refresh, resolving a frustrating autosave bug.
• –**Codex Enhancements**: First-time contributor PRs enable Ask AI to run in non-git directories and output the review URL to the terminal when hosted on Codex Desktop.