Mahal Systems EnterpriseClawBench
Fits teams who have their own agent logs and want to turn them into evals. Borrow the construction pipeline (fixtures recovered, prompts rewritten, rubrics and rules generated) and the harness-plus-model reporting.
What it asks the agent to do
- reproduce a real workplace agent session from its recovered files
- read mixed files and deliver a business artifact (report, sheet, deck)
- call tools in a sandbox to finish the job
What to borrow from it
privacy patterntaxonomygrading design
Caveats
The task data itself is private; only the protocol and one sanitized example are public; judge-graded.
Details
| Workloads | documents, data analysis, workflow automation, general assistant |
|---|---|
| Interaction | single turn |
| Environment | containers |
| End state checked | No |
| Tools | sandbox with files and tools; bring your own harness |
| Grading | llm judge, programmatic check |
| Size | 852 |
| Data | real scraped |
| Personal data | possible |
| License | unknown |
| Harness | own repo |
| Status | active |