Where are the review comments and TODOs stored?

Delve uses a tiered approach: it first attempts to use built-in task tools or external trackers, falling back to a session JSON file or direct output in the conversation.

Can I resume an interrupted review session later?

Yes, Delve persists session state including your current position and the chosen baseline, allowing you to pick up exactly where you left off in a previous session.

How does Delve decide the order of code changes during a review?

Delve builds a symbol graph of your changes to ensure you see function definitions and utility helpers before the code that calls them, providing better context for the review.

What happens if a code change is too large for the screen?

Delve targets a ~40-line chunk size for readability. It automatically groups or splits hunks to maintain high cohesion while ensuring the diff fits comfortably in your terminal.

Delve Interactive Code Review

Name: Delve Interactive Code Review
Author: MattKotsenas

byMattKotsenas

0•

协作工具

Orchestrates interactive, chunked diff reviews that transform reviewer feedback into actionable tasks for downstream agents.

Delve is a specialized code review skill for Claude Code that optimizes the human-in-the-loop experience by breaking down complex git diffs into logical, screen-sized chunks. Using a symbol-aware heuristic, it builds a dependency graph to ensure reviewers see function implementations before their call sites, providing maximum context for every change. Delve streamlines the transition from review to action by capturing feedback as structured TODOs with stable content anchors, allowing downstream AI agents or developers to implement requested changes without losing track of specific requirements.

主要功能

01Intelligent diff chunking based on screen-fit and symbol cohesion

02Interactive terminal-based navigation with integrated feedback capture

03Session persistence to resume reviews and track progress across sessions

040 GitHub stars

05Structured TODO generation with stable, content-based anchors

06Dependency-aware ordering to present implementations before call sites

使用场景

01Performing incremental reviews of uncommitted changes during active development

02Reviewing large feature branches before merging into a target branch

03Generating structured task lists from human feedback for automated implementation

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add mattkotsenas/agent-plugins delve

For use in Claude.ai and ChatGPT

Download Skill