Reviewers see Tracepipe as a practical, easy-to-use way to turn browser-based workflows into reusable AI instructions without hand-writing SKILL.md files. They repeatedly praise the fast trace-to-skill flow, intuitive setup, token cost estimates, and portability across tools like Claude Code, Cursor, and Codex. Several say it helps automate routine work, including Jira processes, and reduces manual documentation. The main criticism is that people want a clearer preview of the experience before trying it, especially a short demo plus more guidance for complex workflows, editing, branching, and updates.
Hey, Max here, co-founder of Tracepipe.
The idea is pretty simple. You do a task once in Chrome the way you normally would, Tracepipe watches what the browser does, and you get a skill your AI agent can use from then on. Works with Claude Code, Cursor, Codex, anything that reads a SKILL.md.
I built it because every company I talk to has bought Claude or Cursor seats and then the agent just kind of sits there, because nobody ever wrote down how the work actually gets done. The first one I made for myself was a refinance order on a lending platform, fourteen steps across three tabs. Took about four minutes to walk through and I haven't done it by hand since.
A few things people usually ask about. It only sees your browser, not your desktop, and passwords never make it into the capture. You see a token estimate before anything uploads and nothing gets charged until you say yes. Every skill can also generate an SOP, basically the same steps as a PDF for the people on your team who aren't using an agent. And the skill is just a folder with a SKILL.md in it, so if a provider changes their pricing on you, you take your skills and leave.
There's a free tier, go try it on something real.
I'd like to hear what you'd point it at first! Skill distribution is a known gap we are working on as our top priority.