Solution

Coding agent training data

Issue-to-fix histories, pull request review threads, and CI outcomes for training and evaluating AI coding agents, licensed with a documented rights position.

The problem

Why this is hard

Agents need to learn the whole change loop — understanding an issue, locating code, editing, running tests, and responding to review — but most corpora contain only the final code, not the process.

What we source

Data types

  • Issue tracker items linked to the commits and pull requests that resolved them
  • Code review conversations and requested changes
  • CI/test outcomes before and after each change
  • Reproducible environments where the supplier can license them

What you receive

Deliverables

  • Linked records (issue → patch → review → outcome) in a documented schema
  • Executable validation notes where tests can be run
  • Rights record and permitted-use statement per source
  • Sampling plan so you can evaluate before committing

Quality

Checks on every delivery

  • Link integrity between issues, commits, and reviews
  • Test validity checks
  • Personal-data scrubbing in conversations
  • Rights and contributor-agreement review

FAQ

Coding agent training data: common questions

Do the tasks include runnable tests?
Where the supplier can license the environment and the tests run, we validate them and report the pass rate. Where they cannot, the dataset card says so.

Tell us what your models need to learn.

Share what you need or what you own. We reply with next steps — or tell you honestly if we can’t help.