Skip to main content
Back to timeline
backnotprop/plannotatorSource publication:

Plannotator moves agent review into a local browser: plans, diffs and terminal replies can be annotated and sent back

Synopsis

Plannotator is a local, browser-based review surface for AI coding agents including Claude Code, Codex, Copilot CLI and Gemini CLI. When an agent proposes a decision, plan or HTML artifact, or finishes work, the app opens in the browser so the user can mark it up and send feedback straight back to the agent to act on. The new Plannotator Inbox gathers messages, questions, files and guided reviews from multiple agents into one local window; answering does not hold a session open, and the reply returns to the session that asked.

Source-provided article image: backnotprop / plannotator
GitHub Trending

Interpretation

AI coding agent output gains a local review step: plans, specs, Markdown, HTML and other interactive artifacts, along with code diffs and pull requests, can be annotated point by point in the browser, and the annotations go back to the agent as feedback. Plannotator positions itself as the missing review layer between the user and multi-agent work, surfacing decisions, prototypes, variations and plans together. The app plugs into agents through plugins, hooks and commands; when an agent proposes a decision, plan or HTML, or finishes work, the app opens in the browser for the user to mark up and comment.

The new Plannotator Inbox is one local window where agents leave messages, questions, files and guided reviews; the user answers a question with a click or annotates a file, then presses Send. Inbox is positioned as the missing review layer between the user and multi-agent work, surfacing decisions, prototypes, variations and plans together. Claude Code, Pi and OpenCode connect through Plannotator's plugins and receive replies as their next turn; other agents reach it over MCP with plannotator inbox mcp.

Herdr Annotate brings the same review style to the terminal: terminal text can be annotated, whole Markdown documents and coding-agent replies reviewed, and the feedback sent straight back to the agent as its next message. Terminal review and in-app review share the same annotation store, so the two accumulate together. Annotations are saved in the Plannotator data directory, so terminal reviews and app reviews compound; the feature works with replies from Claude Code, Codex, Pi, Copilot CLI and Droid.

Perspective

The demonstrated scope is a local, browser-based review surface and its terminal version, integrated through plugins, hooks and commands, with other agents connecting over MCP; in shared links, smaller content lives in the link itself, while larger content is encrypted before upload and expires after 7 days, and anyone with the full link can read it.

The material gives no quantitative result for how the review step affects agent output quality or the number of round trips; in sharing, only smaller content lives in the link itself, while larger content is encrypted before upload, expires after 7 days, and is readable by anyone holding the full link.