handshake-ai/atlas-finance
Finance
Professional
100 expert-level finance tasks in 13 realistic firm environments, served over MCP (Outlook, Slack, Drive, Calendar, OneNote) and graded by an agent-as-judge against weighted, section-gated rubrics.
Run this task
CLI:
inspect eval inspect_harbor/handshake_ai_atlas_finance --model openai/gpt-5Python:
from inspect_ai import eval
from inspect_harbor import handshake_ai_atlas_finance
eval(handshake_ai_atlas_finance(), model="openai/gpt-5")Dataset information
| Harbor registry | handshake-ai/atlas-finance |
| Inspect task | handshake_ai_atlas_finance |
| Latest digest | sha256:4af61568b332d45aa03f0052b85e9a2610a3c706346c17b37c611cbb6787ed23 |
| Samples | 100 |
| Source | https://github.com/Handshake-AI-Research/ATLAS-Finance |
See Task Parameters for the parameter set shared across all Harbor tasks.