🧠 Human-in-the-loop review UI for autonomous AI agents
AgentClick is a skill-based plugin that adds a browser review layer to terminal agents.
Agents propose actions, users inspect and edit in a browser UI, then the agent continues.
Most agents still interact like this:
User → Terminal → Agent → Action
AgentClick adds a review step:
Agent proposes → Browser UI opens → User inspects and edits → Agent executes
This keeps the speed of autonomous agents while adding a human-in-the-loop safeguard before irreversible work.
AgentClick works as a skill / plugin for modern AI agents. Any agent that can run local tools, send HTTP requests, and follow skill instructions can integrate with it.
- Claude Code
- Codex
- OpenClaw
- Custom tool-calling agents
AgentClick extends the agent interaction into a browser UI for tasks like:
- 📧 Email Review — browse inbox, edit drafts paragraph-by-paragraph, preference learning
- 📋 Plan Review — inspect multi-step plans, inject constraints before execution
- 🖥️ Code Review — GitHub-style diff viewer before any script runs
- ✅ Action Approval — approve or reject external actions with reasons
- 🔍 Trajectory Inspection — replay every decision, tool call, and error in a DAG
- 🗳️ Selection Review — agent presents options, you choose
- 🧠 Memory Management — control what the agent remembers across sessions
The goal: keep the speed of terminal agents, but add a real review layer before the agent commits to irreversible work.
- Edit before execution — change the draft, command, or payload instead of only approve/reject.
- Shared visual context — move from raw terminal text to a purpose-built UI.
- Preference learning — feedback from review is persisted so the agent improves over time.
- Framework-agnostic — anything that can
POSTJSON and poll an HTTP endpoint can use it.
npm install -g @harvenstar/agentclick
agentclickThen open http://localhost:38173 in your browser.
For remote access from another device:
agentclick --remote--remote automatically downloads and starts a Cloudflare Quick Tunnel. It prints a public HTTPS URL you can open on your phone or another machine.
Option A — npm global install (AgentClick already running)
Start AgentClick first (agentclick), then tell your agent:
AgentClick is running at http://localhost:38173. Use it whenever you need a browser review UI — for email triage, risky commands, plans, or memory review.
Option B — working inside the AgentClick repo
If your agent can read local files, point it at the skill router:
Load SKILL.md from this repo, start AgentClick locally, and use it whenever you need a browser review UI instead of only terminal output.
The root SKILL.md routes to the right sub-skill automatically.
For OpenClaw in particular, use a stronger model with solid instruction-following, since the workflow is skill-based and depends on the agent following routing instructions reliably.
AgentClick is built around a skill-based architecture. The root SKILL.md acts as a router that directs the agent to the appropriate sub-skill.
| Skill | Path | Purpose | How to Use |
|---|---|---|---|
| Router | SKILL.md |
Entry point that routes the agent to the right review workflow. | Load SKILL.md and use AgentClick UI for review. |
| Action Approval | skills/clickui-approve/ |
Approve or reject risky actions before execution. | Before deleting those files, show me an approval review in AgentClick UI. |
| Code Review | skills/clickui-code/ |
Review shell commands, diffs, and code-related actions. | Show me a code review in AgentClick UI before running that command. |
| Email Review | skills/clickui-email/ |
Review inbox items, drafts, replies, and live email sessions. | Open my inbox in AgentClick UI and let me triage emails. |
| Plan Review | skills/clickui-plan/ |
Inspect and revise proposed plans before the agent runs them. | Show me the plan in AgentClick UI before you start. |
| Trajectory Review | skills/clickui-trajectory/ |
Review multi-step runs, mistakes, and resume points. | Show me what you just did in AgentClick UI so I can review the steps. |
| Selection Review | skills/clickui-selection/ |
Present options to the user and capture their choice. | Show me the options in AgentClick UI and let me choose. |
| Memory Review | skills/clickui-memory/ |
Review memory files and memory-management changes. | Open memory management in AgentClick UI and let me pick which files to include. |
In most cases, telling the agent to load the root skill is enough.
git clone https://github.com/agentlayer-io/AgentClick.git
cd AgentClick
npm install
npm run devDevelopment mode:
- server:
http://localhost:38173 - web:
http://localhost:5173
Production-style single-port serving:
npm run build
npm start