Coding traces
Real coding sessions,
every step kept.
Developer-agent sessions from Cursor, Claude Code and Codex: the prompt, the tool calls, the edits and what happened. Sold by the developers who ran them.
Benchmarks show a model solving tidy tasks. Traces show it working next to a real developer on a real repo.
Every session starts with a person who had a job to do. They prompted an agent, watched it read files and run tests, corrected it, and accepted or threw away the result. That full loop is what we deliver, matched to your training or evaluation brief.
- Tools. Cursor, Claude Code and Codex.
- Fields. Prompt, tool calls, edits, outcomes, task and model metadata, where available.
- Source. Developers who opt in through Cookiy Earn and our open-source Sell Sessions Skill.
One session
From prompt to outcome.
Intent
The prompt and repository context show what the developer was trying to do.
Actions
Tool calls and edits show how the session moved, including the corrections.
Outcome
Accepted, reverted or abandoned, with task and model metadata to filter by.
Opt-in supply
Sold by the developers who ran them.
Our open-source Sell Sessions Skill lets developers review and submit the sessions they already ran, and get paid through Cookiy Earn on every sale. It runs locally, the developer chooses every session, and secrets are scrubbed before anything leaves the machine.
Questions teams ask.
Are these simulated coding sessions?
No. These are real developer-agent sessions. A session contains model-generated content because it records a person actually using an AI tool on their own work.
Is every field in every session?
No. Prompts, tool calls, edits, outcomes and metadata are included where the tool and export provide them. Required fields are checked against a sample before batch delivery.
What goes into a coding brief?
The tools and models you need, the kind of work (languages, repository size, task type) and the fields you require. We pull a sample so you can check real fields before you commit.
How do you handle secrets and personal data?
The Sell Sessions Skill scrubs secrets and PII on the developer's machine before upload. The developer picks every session that leaves the machine.
More human data
Four more kinds of human data.
Speech and conversation
Two-speaker and group recordings with overlap, pauses and barge-ins, plus labels for tone, emotion and delivery.
Learn more →Expert judgment
Doctors, lawyers, teachers, engineers and hundreds of other professions answering domain tasks and explaining their decisions.
Learn more →Private records
Consumer chats, AI histories, photos and purchases, and expert notes, contributed with consent and profile context.
Learn more →Enterprise data
Non-public emails, chats, documents, meetings and private codebases from consenting companies, sourced against a brief.
Learn more →License
Existing recordings and records, with source context, metadata and usage rights.
Commission
New conversations, tasks and recordings from the people your brief describes.
Label and review
Experts annotate and grade against your rubric, with review notes.
Agree the spec and acceptance criteria → review a sample → scale.
Which tools, which tasks?
We will pull a sample.
Video on this site is licensed stock footage of real people, used to illustrate the kinds of data we collect.