SWE-gen-fn (work in progress)

1000 functional programming tasks generated from Y open-source GitHub repos using SWE-gen.

Each task

• is a merged GitHub PR with 1-10 source files edited
• has Fail-to-Pass unit tests
• passes NOP (baseline fails) and Oracle (fix succeeds) validation

Getting Started

Install Harbor:

uv tool install harbor

Run with Codex:

export OPENAI_API_KEY=<YOUR-KEY> 
harbor run --dataset swe-gen-fun \
   --agent codex \
   --model openai/gpt-5.3-codex \
   --n-concurrent 4

This command automatically downloads the tasks.