Browse the corpus of repositories our agent pipeline has dockerized — read the Dockerfiles it produced, replay the agent transcripts, and study where it succeeded and where it gave up.
01
A repo catalog
530 unique repositories the agent has attempted, grouped by language with success rates and median build times.
02
Per-run detail
For every run: the generated Dockerfile, the iteration tree, every tool call the agent made, and whether the smoke test passed.
03
Failure analysis
See where the agent retried, what it learned between iterations, and the build error that ended runs that never went green.