shivam-gupta-ai/repogym-task-validation

RepoGym: inspect the task before the agent

A free static evidence viewer for RepoGym, an MIT-licensed Python library for defining and validating repository-based coding tasks.

What is here

• —Results from a local run of all five bundled tasks on September 22, 2026.
• —A browsable HTML report and downloadable run JSON.
• —Source-install instructions to reproduce validation and reference-solution runs.

All five tasks passed validation with two repetitions. All five reference solutions scored 1.0. These are known-solution controls, not model performance measurements. The Space does not execute submitted code. Local execution is intended for trusted task code; see the source repository's sandbox documentation before using external candidates.

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

Documentation | Source and issues | Shivam's portfolio | LinkedIn

Created by Shivam Gupta. This viewer and RepoGym use the MIT license. Optional model providers have their own billing.