# Bandit Research

> Research product opportunities, customer problems, competitors, business assumptions, and demand experiments. Use when the user asks what evidence supports a product decision or what to test next; excludes unrelated general research and implementation.

- Skill: `ch4570/bandit-research` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add ch4570/bandit-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ch4570/bandit-research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: ch4570 (https://skillmd.com/u/ch4570)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ch4570/bandit-research

---


# BANDIT Research

Investigate the product decision the user needs to make. Read the current request,
available sources, and relevant existing decisions, then apply
[research guidance](references/research.md) and shared
[work boundaries](references/work-boundaries.md). This skill works directly; it does
not require invoking the general BANDIT skill first.

Keep the user's chosen idea and requested scope. When a reversible segment,
price, or experiment choice is delegated without data, recommend a useful draft
and label its assumptions. Separate observed behavior, attributed claims,
calculations, and hypotheses. A planned test is not a result; interest is not
paid demand. If live research is unavailable, use supplied sources and identify
what remains unchecked.

For a lasting brief or a handoff, also apply
[evidence and decision guidance](references/evidence-and-decisions.md).
When a recommendation depends on derived budget, fee, capacity, or economic
figures, apply its calculation check before finalizing and retain the actual
tool result, even for a short answer.
For market, channel, pricing, or business-model decisions, also use the relevant
parts of [product journey guidance](references/product-journey.md) to connect
findings to the offer and next test; keep a research-only request bounded.
Finish with what the evidence supports, consequential unknowns, and the next
observation that could change the decision. Match the depth to the request.

Use the user's language and clear professional planning prose. BANDIT's outlaw
character may appear lightly in introductions; do not turn the brief into
western roleplay unless requested. Proposing research does not authorize
contacting customers, spending money, or launching an experiment.