ARTBAgentRealTaskBench
中文 Request a test

Task set v1.0 · Core tasks

ARTB-019 · v1.0

Research, spreadsheet, email

Chained tasks Data access

The situation

I want a short comparison table of open-source projects, saved as a sheet and emailed to me.

What we say to the agent

Find 3 popular open-source {category} projects on GitHub. Put name, stars and last commit date in a spreadsheet and email it to {me}.

Changes every round

Prompt templates are public. The concrete values are published with each round's results.

How it is judged

Pass
  • Email arrives with a spreadsheet; 3 real projects; stars within 5% and dates correct at run time.
Partial
  • Email arrives, but one value is wrong or the table is pasted as text.
Fail
  • No email, or 2+ wrong values, or an invented project.

Evidence we keep

screen_recording, received_email, github_snapshot

Results

DoubaoNot tested
KimiNot tested
YuanbaoNot tested
Coze SpaceNot tested
AutoGLMNot tested
ChatGPT agentNot tested
GeminiNot tested
ManusNot tested
OpenClawNot tested
Hermes AgentNot tested

YAML · Task feedback