decision-lab (dlab)

dlab runs autonomous coding agents in frozen Docker environments with domain-specific skills and parallel subagents. You package the environment, prompts, and skills into a decision-pack, point it at data, and get back reports and recommendations that hold up to scrutiny.

When to use this skill

• Creating a new decision-pack (interactive or programmatic)
• Designing agent system prompts for data science workflows
• Understanding how parallel agents, consolidators, and retry protocols work
• Analyzing a completed session's logs, outputs, and artifacts
• Running or configuring dlab CLI commands

Workflow overview

1. Create a decision-pack — scaffold with dlab create-dpack wizard or generate_dpack() programmatically
2. Design agents — write orchestrator, subagent, and parallel agent configs
3. Run a session — dlab --dpack <path> --data <data> --prompt "..."
4. Monitor — dlab connect <work-dir> (live TUI) or dlab timeline <work-dir> (Gantt chart)
5. Analyze results — browse session directory, logs, parallel instance outputs