I built AgentClick because we kept getting burned by AI agents acting too fast.
The pattern is always the same: you ask Claude Code or Codex to do something, it generates a response, you get a y/n prompt in the terminal, and you approve it without really reading it. Then it sends the wrong email, runs a destructive command, or commits to a plan you'd never agree to if you actually saw it laid out.
The core problem: terminal y/n is not a real review step.
AgentClick adds a browser-based review layer between "agent proposes" and "agent executes." The agent drafts something, a UI opens in your browser, you can actually read, edit, and approve it visually — then the agent continues.
It works as a skill/plugin for: - Claude Code - Codex - OpenClaw - Any agent that can call HTTP tools
What you can review: - Email drafts and inbox triage - Shell commands before execution - Multi-step plans - Memory updates
The key difference from just adding a confirmation step: you can actually edit the content, not just approve/reject. Change the tone of an email, fix a command flag, remove a step from a plan — then let the agent proceed.
Install: `npm install -g @harvenstar/agentclick`
Works locally on localhost, or use `--remote` for a Cloudflare tunnel to review from your phone.
I built AgentClick because I kept getting burned by AI agents acting too fast.
The pattern is always the same: you ask Claude Code or Codex to do something, it generates a response, you get a y/n prompt in the terminal, and you approve it without really reading it. Then it sends the wrong email, runs a destructive command, or commits to a plan you'd never agree to if you actually saw it laid out.
The core problem: terminal y/n is not a real review step.
AgentClick adds a browser-based review layer between "agent proposes" and "agent executes." The agent drafts something, a UI opens in your browser, you can actually read, edit, and approve it visually — then the agent continues.
It works as a skill/plugin for:
- Claude Code
- Codex
- OpenClaw
- Any agent that can call HTTP tools
What you can review:
- Email drafts and inbox triage
- Shell commands before execution
- Multi-step plans
- Memory updates
The key difference from just adding a confirmation step: you can actually edit the content, not just approve/reject. Change the tone of an email, fix a command flag, remove a step from a plan — then let the agent proceed.
Install: `npm install -g @harvenstar/agentclick`
Works locally on localhost, or use `--remote` for a Cloudflare tunnel to review from your phone.
I've been building a review UI layer for coding agents (Claude Code, Codex) that lets you actually inspect and edit what the agent is about to do before it executes: https://github.com/agentlayer-io/AgentClick
Turns out most of the "dumb" mistakes OP is talking about are catchable — you just need to actually see them before they ship.
This resonates. The problem I keep running into isn't that the model is bad — it's that the feedback loop is too thin. A y/n in the terminal isn't enough to catch when the model does something subtly wrong.
Cool project — been looking for something like this.
Just opened a PR with a couple of new macOS actions (empty_trash + toggle_do_not_disturb). Happy to contribute more and quick chat if you're open to it.
Maybe a middle ground — let people pick a pseudonym and a one-line tagline, but no real identity. You keep the warmth of anonymity while still letting listeners find voices they connected with.