lgtm

The lgtm (Looks Good To Me) skill provides an interactive terminal user interface (TUI) designed for structured code and document reviews within a terminal environment. It is particularly effective for reviewing AI-generated plans, technical documentation, or code proposals that require precise, line-by-line feedback. By integrating directly into the Claude Code development cycle, this skill allows users to maintain a side-by-side workflow in tmux, making it an essential tool for developers and technical writers who need to iterate on text-based assets while maintaining context.

• Line-by-line navigation and commenting using six distinct categories: Blocker, Concern, Question, Suggestion, Praise, and Acknowledge.

• Integrated syntax highlighting for code blocks within markdown documents to ensure readability during review.

• Real-time feedback export capability in Markdown or JSON format, enabling seamless transition from TUI review to actionable project updates.

• Support for session persistence, allowing reviewers to save, load, and manage multiple feedback sessions using JSON storage files.

• Automatic tmux integration for side-by-side review workflows, ensuring that Claude Code can respond to feedback immediately upon the user quitting the TUI.

• Context-aware reviews that allow Claude to load its own previous responses or active planning documents directly into the TUI for immediate critique.

• Usage requires the Bun runtime (v1.0.0+) and ideally the tmux terminal multiplexer for the best side-by-side experience.

• The tool is invoked via the /lgtm command, followed by a file path or the --stdin flag for piping content.

• When a review session concludes, the skill automatically exports the gathered feedback, which the agent then reads and processes to implement the requested changes or address concerns.

• Users should focus on prioritizing 'Blocker' and 'Concern' type comments to facilitate the fastest path toward project resolution and iterative improvement.

• Files are handled via absolute paths, and the tool uses the Read tool to maintain full context, ensuring that line-level comments map accurately back to the original source material.