com um clique
live-kbench
live-kbench contém 4 skills coletadas de ARiSE-Lab, com cobertura ocupacional por repositório e páginas de detalhe dentro do site.
Skills neste repositório
Import or export live-kBench kArena bugs and dataset definitions with HuggingFace Hub. Use when Codex is asked to sync, publish, push, pull, import, export, mirror, snapshot, or restore kArena datasets through kArena.hf_sync, HFBugExporter, HFBugImporter, export_to_hf, import_from_hf, or HF_TOKEN.
Generate ready-to-paste kArena shell task factory snippets for live-kBench operations. Use when Codex is asked to start, deploy, run, poll, bootstrap, sync, or manage kArena workflows in karena-shell, including crawl_bugs, submit_kcache, poll_kcache, build_kenv_images, crash evals, LLM judge evals, localization, developer patch population, and HuggingFace import/export tasks.
Write or update live-kBench kEnv agent integrations. Use when Codex needs to create a kenv-invoker.py, adapt an agent to the kEnv protocol, add a docker/kenv-*.Dockerfile, debug kEnv container inputs or outputs, or ensure an agent writes patch.txt, traj.json, and log.txt for live-kBench evaluation.
Deploy, upgrade, shut down, or troubleshoot kGymSuite for live-kBench in local Docker Compose or GCP/remote multi-server environments. Use when Codex is asked to deploy kGym, start local kGym services, configure deployment/local or deployment/gcp, run kgym.py new-deploy/upgrade/down/config-artifact-reg, prepare kGym storage, or verify scheduler/dashboard/worker services.