what is senior-swe-bench-v2026.06 fr?

This repository holds the dataset used by Senior SWE-Bench, a benchmark from Snorkel AI that tests how well AI coding agents perform on real software engineering tasks. The README itself is short and mainly points to a separate website and documentation for the actual details of what the benchmark measures and how the tasks were built. What this repo provides is the set of public test tasks in a format readable by a tool called Harbor, an evaluation framework also made by the same team. Because Harbor can read straight from this repository, someone wanting to run the benchmark does not need to download the dataset separately, only install Harbor itself. To actually run an evaluation, a user sets an API key for whichever AI model they want to test, such as Anthropic or OpenAI, picks an agent framework and a specific model, and then runs a single Harbor command pointing at this repository. The README shows both running it through something called the Harbor Hub and running it by pointing Harbor directly at the GitHub repo. It also lists a few optional environment variables for overriding which model is used for grading and classifying the agent's attempts, but does not explain the grading process in the README itself. Overall, this repo is a data and configuration companion to the Harbor benchmarking tool rather than a standalone application, and the README is sparse, so most of the actual explanation of how the benchmark works lives on the linked website and documentation pages instead of here.