Plannotator brings browser-based annotation to agent plans and code diffs
Plannotator is an open-source tool that lets developers annotate AI agent plans, code diffs, and HTML artifacts in the browser, feeding markup back into the agent session, with support for nine agent harnesses.
Score breakdown
Plannotator replaces terminal-based plan approval with a structured, browser-based review layer that feeds annotations directly back into agent sessions, addressing the human-review bottleneck the post identifies as the limiting factor as agents become more capable.
- 01Originally built as a hook for Claude Code's plan mode, opening plans in the browser instead of the terminal
- 02Annotations made in the browser feed back into the agent session
- 03Supports PR-style code review for git, jj, p4, GitHub, and GitLab
Plannotator began in January as a hook into Claude Code's plan mode, replacing terminal-based plan approval with a browser interface where developers can mark up the plan like a document and send annotations back into the agent session. From that starting point, the tool expanded into a broader review layer for agentic workflows, adding PR-style code review across git, jj, p4, GitHub, and GitLab, as well as annotation support for markdown, URLs, folders, and the HTML artifacts that agents increasingly produce.
The author frames the tool around the observation that as agents improve, the bottleneck shifts to human review before and after agents work.
The tool now supports nine agent harnesses — Claude Code, Codex, Copilot CLI, Gemini, OpenCode, Kiro, Droid, Amp, and Pi — and has grown to over 100 contributors. Sharing is opt-in and end-to-end encrypted: plans are encrypted in the browser, the server stores only ciphertext, and the decryption key lives in the URL fragment, meaning nothing leaves the user's machine unless they choose to share. The project is released under MIT or Apache-2 licenses. The author frames the tool around the observation that as agents improve, the bottleneck shifts to human review before and after agents work.
Key facts
- 01Originally built as a hook for Claude Code's plan mode, opening plans in the browser instead of the terminal
- 02Annotations made in the browser feed back into the agent session
- 03Supports PR-style code review for git, jj, p4, GitHub, and GitLab
- 04Also annotates markdown, URLs, folders, and HTML artifacts produced by agents
- 05Nine agent harnesses supported: Claude Code, Codex, Copilot CLI, Gemini, OpenCode, Kiro, Droid, Amp, and Pi
- 06Over 100 contributors; open source under MIT or Apache-2 licenses
- 07Sharing is E2E encrypted: plans encrypted in the browser, server stores ciphertext, key lives in the URL fragment
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 12, 2026 · 10:05 UTC. How this works →