Rank #05 · Weekly · Agent-Skill
Plannotator: A Human-in-the-Loop Review Workspace for Coding Agents
When an agent proposes a plan, finishes a change, or leaves you a question, open a local browser review surface, mark it up, and send the feedback back to that session.
What it does
Plannotator is a local, browser-based review surface for AI coding agents. GitHub description: "The workspace to stay in the loop and on top of agents - shape, decide, plan, annotate, review code, share with team, and send all feedback to agents with one click." Site: plannotator.ai.
README claims (as verified):
- Annotate plans, specs, messages, Markdown, HTML, and other artifacts, then send the comments to your agent. You can point it at a local file or folder, fetch a URL, render HTML as-is, or reopen the agent's last message (
/plannotator-annotate,/plannotator-last). - Code review of local changes or remote PRs/MRs: comment on diffs, suggest code, and return the comments to the agent. The README lists Git, GitButler, Jujutsu (
jj), Perforce (p4), GitHub, and GitLab. Commands include/plannotator-review(uncommitted changes or a PR/MR URL) andplannotator review --patch-file. - Plan mode is wired through each harness's hooks — when the agent creates a plan, the Markdown review surface opens. Approve to let the agent proceed; deny to send structured feedback. A later plan diff shows what changed.
- Plannotator Inbox (
plannotator inbox) is one local window where agents leave messages, questions, files, and guided reviews while they keep working. Replies go back to the session that asked. Claude Code, Pi, and OpenCode connect through plugins; other agents can reach it over MCP (plannotator inbox mcp). - Listed coding-agent integrations: Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness, and Pi. The Inbox README also mentions personal assistants (Grok Bot, Muse, Hermes, OpenClaw). A VS Code extension, plus optional Obsidian / Bear / Notion plan-saving, is documented separately.
- Installer (macOS / Linux / WSL:
curl -fsSL https://plannotator.ai/install.sh | bash; Windows PowerShell script also documented) installs theplannotatorbinary and can configure hooks, skills, and slash commands for detected agents.--minimalinstalls only the binary. Herdr Annotate and Plannotator TUI are related terminal surfaces, not this repo's main app. - Privacy (README): the project says it does not collect usage telemetry or analytics; plans, diffs, annotations, drafts, history, and configuration stay local by default. Surfaces do check GitHub for the latest release when they load (no plan/review content; no opt-out at the time of writing). Content leaves the machine when a feature needs it (URL fetch, GitHub/GitLab review, Ask AI / review agents, link sharing). Workspaces is a separate hosted product and does not inherit the open-source app's local-storage model. Legacy link sharing is described as moving toward deprecated support.
License: Apache-2.0 OR MIT at your option (README wording; root files LICENSE-APACHE and LICENSE-MIT). Contributions are dual-licensed under the same terms unless you state otherwise. GitHub's license API currently shows only Apache-2.0 — do not treat that badge as a single-license project.
Who it's for
- Developers who want a human-in-the-loop review layer on plans, artifacts, and diffs from coding agents they already use.
- Teams that want comments and approvals to return to the same agent session, rather than living only in a chat transcript or a PR UI.
- People who prefer a local-first review surface, with optional sharing and a separately hosted Workspaces product if they need team collaboration later.
Less ideal if you only need a standard GitHub/GitLab review UI, do not run one of the supported agent harnesses (or MCP), or want a fully offline tool with no release-check or optional network features. Remote / SSH / devcontainer mode listens on every interface with no login — keep that port closed to networks you do not trust.
Stars / language / license
13,664 stars · TypeScript · Apache-2.0 OR MIT
Data verified: 2026-11-20 09:04 Asia/Shanghai (official Daily Top10 gate; language confirmed via GitHub API get_repository; dual license confirmed via README + LICENSE-APACHE + LICENSE-MIT)
Trending note: +3,468/wk (weekly Trending window; selection captured 09:04 Asia/Shanghai)
About the license: this repository is dual-licensed. You may choose Apache-2.0 or MIT. This is a summary, not legal advice.
Category tags
Agent-Skill · code-review · plan-review · human-in-the-loop · TypeScript · CLI · MCP · local-first
How it differs from similar projects
Compared with reviewing only inside the agent's chat. Plannotator adds a dedicated annotation and diff surface (and an Inbox for async questions) so feedback is structured and sent back to the session that asked.
Compared with GitHub / GitLab review UIs. Those remain the system of record for hosted PRs. Plannotator can open a PR/MR URL, but it is aimed at the agent loop — local diffs, plans, and artifacts — not at replacing your forge.
Compared with Alibaba Open Code Review (#04). That project is a hybrid deterministic + LLM review CLI that generates line-level comments. Plannotator is a human review workspace that lets you annotate and return feedback to an agent. They can sit in the same workflow; TidyHow has not compared them as substitutes.
Caveats. Optional Ask AI / review-agent features send selected context to a configured provider. Workspaces is hosted and separate. Installer and release-check network behavior are documented in the README and privacy policy — read those if that matters to you. No independent productivity benchmarks are claimed here.