Wisp AgentDiff：Claude Code 并行子智能体工作流的代码审查工具

Wisp AgentDiff is an open-source code difference review tool designed specifically for Claude Code's parallel subagent workflows. It addresses the key challenge of reviewing and merging code changes from multiple subagents by providing automatic capture of subagent working directory states, a tabbed terminal user interface (TUI), and flexible change management (approve, revert, merge). This tool helps developers maintain control over AI-generated code in parallel workflows.

With the popularity of AI programming assistants like Claude Code, developers are exploring parallel subagent workflows—where a main agent splits complex tasks into sub-tasks handled by multiple subagents in parallel. However, this paradigm introduces a critical problem: how to effectively review and merge code changes from multiple subagents without losing context or control.

Wisp AgentDiff's core features include:

1. Automatic Working Directory Capture: Tracks each subagent's working directory state automatically, no manual intervention needed, preserving full context for review and traceability.
2. Tabbed TUI Interface: Intuitive terminal interface with multi-tab browsing for independent subagent change views, clear diff displays, and keyboard shortcut support.
3. Flexible Change Management: Supports three key operations—Approve (accept changes), Revert (undo specific subagent modifications), Merge (integrate selected changes into the main branch) for fine-grained control.

Wisp AgentDiff is deeply integrated with Claude Code's workflow: it understands subagent call patterns, automatically identifies multiple working directories from parallel tasks, and seamlessly fits into existing development processes. Its lightweight architecture focuses on a single responsibility (code diff review and merge), minimizes dependencies for easy installation, and prioritizes terminal use for developer convenience.

Wisp AgentDiff adds value in several scenarios:

• Complex Refactoring: Assign different modules to subagents and review each module's changes independently, avoiding "all-or-nothing" decisions.
• Multi-Solution Exploration: Compare parallel implementations from subagents, select the best practices, and isolate failed attempts.
• Team Code Review: Enable parallel reviews by multiple reviewers, independent decision-making for each subagent's changes, and full audit trails for traceability.

• Vs Traditional Git Workflows: Offers finer-grained control (subagent-level tracking vs branches), instant visualization without branch switching, and a more intuitive TUI than command-line operations.
• Vs AI Built-in Features: Complements Claude Code's code generation by managing batch subagent outputs, providing visual diffs, and allowing fine control over each subagent's changes (something built-in tools lack).

Wisp AgentDiff fills an important gap in the AI-assisted programming toolchain by providing a complete code review solution for parallel subagent workflows. It helps developers manage AI-generated code effectively while maintaining quality control. As the subagent programming paradigm matures, tools like Wisp AgentDiff will become increasingly essential to harness AI's potential without compromising code quality.