How to boost an office, web or coding agent on WorkBuddy tasks with harness2

Authorization notice. harness2 is dual-use. Only use it on software, sites and systems you own or have explicit permission to test.

Who it is forBuilder of an internal assistant that edits documents, browses, or writes code and wants a measurable harness upgrade

The problemHand-tuning prompts for office, web and code tasks is guesswork with no consistent score to compare against.

The outcomeA judged WorkBuddy run for one subset showing whether a Propose-Probe-Compose step beat the base harness.

What you need first

• Linux, Python 3.12 and uv for installation
• No credentials needed for the local demo (harness2-demo)
• Model-backed runs need Node.js 20+, the Google Cloud CLI, a Vertex AI project and model IDs in .env, and gcloud application-default login
• LAB needs Pandoc and Poppler; WorkBuddy-Bench needs Docker
• Model-backed benchmark runs incur API costs

Steps

1. Install the benchmark (needs Docker): harness2-setup wb
2. Check prerequisites: harness2-check --bench wb --subset office
3. Refine one task (inferred, adapting the LAB example to WorkBuddy): harness2 --bench wb --subset office --k 1 --mode parallel --max-tasks 1
4. Judge and report with matching flags: harness2-judge --bench wb --subset office --k 1 --mode parallel
5. Read the report: harness2-report --bench wb --subset office --k 1 --mode parallel

Prompt for your agent

In google-research/harness2, set up WorkBuddy-Bench with harness2-setup wb and run one task from --subset office with --k 1 --mode parallel, then judge and report with identical flags. Verify Docker is running first and that the subset flag is passed on every command after setup.

Caveat. WorkBuddy-Bench needs Docker and the --subset flag on every command after setup, and the README guide for WorkBuddy should be checked for the exact task-selection flags before running.

• Markdown version
• JSON version