什麼是 Plannotator

Plannotator is an open-source tool designed to bring human-in-the-loop control to AI coding agents. It acts as an intermediary layer that intercepts agent-generated plans and code changes, providing a dedicated interface for review, annotation, and feedback. Instead of relying on terminal-based outputs, users can annotate plans, mark deletions, and suggest modifications directly. These annotations are converted into structured feedback that the agent can interpret, ensuring the final output aligns with developer intent. With native support for tools like Claude Code, Copilot, and Gemini, Plannotator bridges the gap between agent autonomy and developer oversight, allowing for PR-style diff reviews and iterative refinement before code is ever committed.

Plannotator 的核心功能

Plan Interception

Automatically hooks into agent workflows to pause execution for human review.

Structured Annotation

Allows users to mark deletions, add comments, and suggest replacements on agent plans.

PR-Style Diff Viewer

Provides side-by-side or unified diff views for uncommitted agent code changes.

Direct Feedback Loop

Converts human annotations into machine-readable feedback for the agent to process.

Version History

Tracks every revision and plan diff to maintain a clear audit trail of changes.

Local Execution

Runs locally on your machine to ensure privacy and security for your codebase.

如何使用 Plannotator

Install Plannotator via the provided shell script command.,Continue using your preferred AI coding agent as you normally would.,When the agent proposes a plan, Plannotator automatically intercepts the step.,Review the plan in the dedicated workspace, adding inline comments or marking deletions.,Approve or deny the plan; your structured feedback is sent directly back to the agent.,Run /plannotator-review to inspect agent-written code changes in a PR-style diff viewer before committing.

Plannotator 的使用情境