benchflow/skillsbench
Assistants
SkillsBench: agent benchmark measuring how effectively models compose and use modular skills (folders of instructions, scripts, and resources) to complete specialized workflows spanning science, engineering, and professional domains.
Run this task
CLI:
inspect eval inspect_harbor/benchflow_skillsbench --model openai/gpt-5Python:
from inspect_ai import eval
from inspect_harbor import benchflow_skillsbench
eval(benchflow_skillsbench(), model="openai/gpt-5")Dataset information
| Harbor registry | benchflow/skillsbench |
| Inspect task | benchflow_skillsbench |
| Latest digest | sha256:145925c10bc09425dc0201772cfa50d9b800010081cf5ad77969554a644d7ae1 |
| Samples | 87 |
| Source | https://github.com/benchflow-ai/skillsbench |
See Task Parameters for the parameter set shared across all Harbor tasks.