benchmark-tasks

An offline benchmark task using Terraform, Python, and Graphviz to reconstruct corrupted machine learning model lineage metadata, query Hugging Face dataset split counts, and render color-coded lineage graphs.

An AI coding benchmark task evaluating automated rebase conflict resolution, MLflow tracking security, secret redaction, and Hugging Face model evaluation safety.

An automated media verification & reconciliation engine resolving conflicting SQLite attestations, Ed25519 signed release manifests, and lost Git verifier policies via Git reflog recovery.

Company-neutral workflow kit for creating, reviewing, calibrating, and packaging Harbor / Terminal-Bench tasks with LLM agents.