I Vibecoded E-Signature in an Afternoon

I started the clock at 1:00 on a Tuesday afternoon, with an AI coding agent and the TurboDocx Quickstart Skill open in front of me. By 1:10 there was working code in the repo. At 1:14 I sent myself a real contract, signed it, and watched the completed PDF and audit trail come back. The signature was out the door before my coffee got cold. The integration was the fast part. The careful part, as it should be, was the review and the edge cases that came after.

Here is the real timeline, so you can calibrate your own expectations.

Everything above starts here:

After that, it is plain English in your agent. The full step-by-step is in the Claude Code guide and the Cursor guide, and the deeper walkthrough lives in Vibecode an E-Signature Flow.

The agent wrote the happy path in minutes. What little time remained went to the things that make a feature production-ready, not the SDK wiring:

The timeline above is clean because the work mostly was. But a few moments are worth slowing down on, because they are the ones that actually decide whether you ship by mid-afternoon or burn the whole day.

The first stretch was scoping, not coding. Before I let the agent touch anything, I wrote one sentence describing what finished looked like: a contract goes out, a signature comes back, the signed PDF lands somewhere I control. That sentence is what kept the prompt tight. I was not asking for a signing platform. I was asking for one flow, end to end. Installing the skill and grabbing a key took longer than scoping did, and most of that was the signup email.

The middle stretch was the send flow and the first real email. The agent wrote a single sendSignature call: a file, a recipient, and a signature field. I pointed it at my own inbox first, which is the move I would recommend to anyone. When that request email actually showed up, I opened the link, signed, and the completed PDF came back with an audit trail attached. That is the moment the feature stops being theoretical. It also surfaced the one snag of the day. My very first send threw before any email left the building, and the reason was unglamorous: senderEmail is required, and I had not put TURBODOCX_SENDER_EMAIL in my env file. One line added, one re-run, and the email arrived. If your first send fails fast with nothing in your inbox, check that env var before you check anything else.

The last stretch was the edge cases and shipping. Signed is the easy outcome. The ones that bite later are the recipient who declines and the request that quietly expires, so I wired both into the completion webhook alongside the success path. Then I read the diff like any other pull request, confirmed the API key never left the server, and merged. The afternoon I had set aside was still mostly ahead of me.

If you want the benchmark version of this, with the clock running against a more traditional build, see how long it takes to build an e-signature app. And if you would rather just start, the Quickstart Skill is the same one I used here.