RepoGymInspect the task before the agent

A coding-agent score starts with a task you can trust.

Define a repository snapshot, test expectations and reference solution. Then check the baseline, constraints and repeatability before running an agent. This Space presents evidence from an actual local run of RepoGym's five bundled tasks.

Source & quickstart Documentation Download run JSON

Reference controls, not an AI leaderboard. The golden agent applies known reference solutions. A 100% control score checks the supplied task setup; it does not measure an LLM or establish security against hostile code. This is a static evidence viewer, not a hosted code-execution service.

01 / Define

Five inspectable examples

Python median bug, JavaScript slugify bug, rate limiter feature, deduplication performance and order-pipeline refactor.

02 / Validate

5/5 tasks passed

Validation ran with two repetitions on September 22, 2026. Inspect the task manifests, hidden tests and reference patches in the MIT-licensed source.

03 / Inspect

Open the evidence

The report below contains the reference-solution results. Expand a task to see its score and grading details.

Reference-solution report

Reproduce this run locally

Python 3.9+ and Git. Node.js is needed for the JavaScript example. Run local tasks only when you trust their code.

git clone https://github.com/shi1720/repogym.git
cd repogym
python3 -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"
repogym validate tasks --repeat 2
repogym run tasks --agent golden --report report.html

The tested code revision was 80db121. Documentation corrections were published in ccb9e0d without runtime changes. Timings depend on your environment.