super-lab-lite

name	super-lab-lite
description	Lightweight multi-agent research orchestration using one coordinator, three domain leads, and three lightweight research agents. Use for medium-size research, market scans, competitor comparisons, prior-art style exploration, report planning, and any task that benefits from parallel domain decomposition without running a full high-cost research lab.

Super Lab Lite

Super Lab Lite is a small public orchestration pattern for splitting a research request into three domains, gathering evidence in parallel, and synthesizing the result into one useful answer.

It is designed for practical research tasks where a single model pass is too shallow, but a large multi-agent workflow is too expensive.

When To Use

• Medium-size research questions with 3 to 5 natural subtopics.
• Market scans, competitor comparisons, product research, and technical surveys.
• Report drafts where evidence gathering and synthesis should be separated.
• Public-safe exploratory analysis with a clear audit trail.
• Cost-sensitive work that does not need a full premium-agent stack.

Do not use it for:

• one-off simple questions,
• high-security work with sensitive data,
• tasks requiring strict legal, medical, or financial conclusions,
• research that cannot be decomposed into clear domains.

Public Model Pattern

Use the following roles. The exact model names can be adapted to the user's provider.

Coordinator:
  - breaks the request into three domains
  - assigns domain prompts
  - merges final results
  - resolves conflicts

Domain leads:
  - analyze one domain each
  - request supporting facts or examples
  - write domain reports

Research agents:
  - gather facts, links, snippets, or local-file evidence
  - avoid final judgment
  - return structured notes to domain leads

Workflow

1. Plan

Ask the coordinator to produce:

• the main research question,
• three non-overlapping domains,
• a domain prompt for each lead,
• an evidence request for each research agent,
• the final output format.

2. Gather

Run the three research agents in parallel where possible. Keep their job narrow:

• collect facts,
• inspect files,
• summarize sources,
• list uncertainties,
• return structured notes.

3. Analyze

Each domain lead turns one research packet into:

• domain summary,
• key evidence,
• risks or gaps,
• recommended follow-up.

4. Synthesize

The coordinator merges the domain reports:

• remove duplicates,
• surface disagreements,
• identify missing evidence,
• write the final answer,
• include a short audit note.

Output Structure

Prefer this final shape:

## Executive Summary

## Domain Findings

### Domain 1
### Domain 2
### Domain 3

## Cross-Domain Synthesis

## Gaps and Uncertainties

## Recommended Next Actions

Scripts

• scripts/lite_orchestrator.py: standalone Python skeleton using the Anthropic SDK.
• scripts/claude_code_runner.md: Claude Code / agent-tool usage pattern.

The Python script expects an API key in ANTHROPIC_API_KEY when run directly. Do not hard-code keys in the repo.

Quality Checks

Before returning the final result:

• confirm that all three domain reports were included,
• mark missing or weak evidence,
• separate facts from judgment,
• avoid unsupported certainty,
• include the most useful next action.

Safety Boundary

This public version is a lightweight orchestration recipe. It does not include private scoring rules, protected workflows, tuned internal prompts, private datasets, or proprietary research benchmarks.