hte-cli

Human Time-to-Completion Evaluation CLI - A tool for running assigned cybersecurity tasks with timing and result tracking.

Installation

# Recommended (pipx)
pipx install hte-cli

# Or with pip
pip install hte-cli

Quick Start

1. Login (get credentials from your coordinator):

   hte-cli auth login
   

2. View your assigned tasks:

   hte-cli tasks list
   

3. Run a task:

   hte-cli tasks run
   

Commands

• hte-cli auth login - Authenticate with the API
• hte-cli auth status - Check authentication status
• hte-cli tasks list - List your pending tasks
• hte-cli tasks run [TASK_ID] - Run a task (defaults to highest priority)
• hte-cli tasks pull-images - Pre-pull Docker images for upcoming tasks
• hte-cli version - Show version info

System Requirements

All Platforms

• Python 3.11+
• Docker Engine >= 24.0.6
• Docker Compose >= 2.21.0 (included in Docker Desktop 4.25+)

Windows

• Docker Desktop with WSL2 backend enabled
• WSL2 installed and configured (Microsoft docs)

macOS

• Docker Desktop (Intel or Apple Silicon)
• Note: Apple Silicon (M1/M2/M3) runs x86 containers via emulation - expect slightly slower performance

Linux

• Docker Engine >= 24.0.6 with Compose plugin
• User added to docker group: sudo usermod -aG docker $USER (log out and back in after)

Verify Docker Setup

# Check Docker Engine version (need >= 24.0.6)
docker version --format '{{.Client.Version}}'

# Check Docker Compose version (need >= 2.21.0)
docker compose version --short

# Test container can start
docker run --rm hello-world

If your Docker Compose version is too old, update Docker Desktop or install the latest compose plugin:

# Linux: Update compose plugin
sudo apt-get update && sudo apt-get install docker-compose-plugin

Configuration

Set HTE_API_URL environment variable to use a custom API endpoint:

export HTE_API_URL="http://your-server.com/api/v1/cli"

Support

For issues, contact your study coordinator or open an issue at: https://github.com/sean-peters-au/lyptus-mono

Developer Notes

This CLI is a thin wrapper with no consequential research decisions. It:

• Wraps Inspect AI's human_cli agent for task execution
• Syncs results to the backend API
• Handles authentication via OAuth-style code exchange

The research-relevant code lives elsewhere:

• Task sampling: scripts/sample_tasks_for_trials.py
• Scoring criteria: src/human_ttc_eval/datasets/*/
• Methodology: docs/methodology/human-expert-methodology-guide.md