Agent Tripwire: approval gate for consumer agents driving your browser

Nothing I found matches the exact pitch, a paid consumer extension that works with any agent and holds submits on government, police, bank and card-field pages. Each part already exists somewhere, though. Gödel Labs markets the same idea of pausing payments and form submissions. Claude in Chrome blocks financial sites and purchases outright, so agent makers are covering the riskiest cases themselves. Open-source tools like agent-browser-extension and Conduit already make users approve each site. A newcomer could still win as a consumer safety layer that works with any agent. It would sit outside the AI model, apply one high-risk rule set to Dots, Grok Bot, Muse and open-source agents alike, and show the exact data before it is sent, where existing tools mostly gate which sites an agent can open. The activity log is the less contested part. AI Browser Guard keeps only five sessions and Agent Performance Lab is built for developers, so neither is a lasting, exportable record a freelancer could use when an invoice or a filing goes wrong. The risks are that agent makers keep adding their own blocks, that agents running in their own browser or in the cloud may bypass extensions entirely, and that you would be competing with free tools for a small group of paying prosumers.

• Gödel Labs Secure Browser Agents

  Its marketing describes a browser extension that pauses payments, form submissions and other irreversible clicks for a human, and that can allow, hold or block each page read, form fill, download and click. This is the closest direct competitor to Tripwire. The claims are unverified.

• Claude in Chrome (Anthropic)

  Built into Anthropic's agent. It blocks banking, investment and crypto sites, prohibits purchases and financial transactions whatever the settings, and offers Manual, Auto and Skip permission modes. This shows the agent makers are covering the riskiest cases themselves.

• Melaya

  A Firefox agent extension with user-controlled approvals in a side panel. In its form-filling case the agent prepares the form and the user reviews it before it is submitted.

• agent-browser-extension (open source)

  An open-source Chrome extension and local bridge for AI agents. It starts in an ask-first mode where the user approves each site, and agents cannot grant themselves access.

• Conduit

  A local-first browser control extension for AI agents with explicit per-site permissions.

• AI Browser Guard (OpenA2A)

  Detects when automation tools or agents take over the browser. It keeps an on-device timeline of each agent action with a timestamp, the URL and whether it was allowed or blocked, but only for the last five sessions. It overlaps with the activity log idea.

• Agent Performance Lab

  Records opt-in sessions locally with timeline events and network metadata, flags sensitive data and exports reports. It is aimed at developers and QA teams, not consumers.

• ContextFort

  Reported as a Y Combinator company. An enterprise Chrome extension that detects agent control, records browser actions and blocks risky behavior, sold to security teams.

• Oso for Agents browser extension

  An enterprise extension, available only through managed deployment, that captures prompts and tool calls for ChatGPT, Claude.ai and Gemini.

• Okolos

  A pre-alpha open-source consumer browser security project that strips hidden instructions aimed at AI assistants from pages before the assistant reads them. It is a nearby consumer agent-safety effort.

Checked 2026-10-10