Clean-Room Risk Screen

Overview

This skill checks whether a proposed path relies on employer-confidential materials, sensitive data, restricted customer access, or unsafe regulated-role framing. The goal is not to provide legal advice. The goal is to separate allowed behavior from risky behavior and propose safer adjacent paths.

Load references/risk-categories.md when classifying risk. Load references/safer-alternatives.md when the original path is unsafe.

When to use

• When a user wants to monetize domain knowledge tied to a job or employer
• When an offer may drift into legal, compliance, healthcare, tax, or similar regulated-role overreach
• When a product plan may depend on PHI, employer data, customer data, or unclear IP ownership

Do NOT use when:

• There is no meaningful IP, data, or role-boundary risk in the plan

Workflow

1. Identify risk category

Classify risk across:

• employer IP/confidentiality
• customer access rights
• sensitive or regulated data
• licensed or regulated authority
• unsafe claims or positioning

2. Decide allowed, conditional, and blocked actions

Produce three buckets:

• allowed
• conditional
• blocked

Be concrete. Do not hide behind vague warnings.

Do not allow paraphrasing or recreating employer-confidential materials from memory as a fake clean-room workaround.

3. Propose a clean-room path

If the original path is risky, propose a safer adjacent route based on:

• public data
• independent firsthand knowledge
• newly created original assets
• generalized workflow understanding
• human-in-the-loop support instead of delegated judgment

4. Produce the memo

Return both:

• a readable risk memo
• a JSON block using assets/risk-screen-template.json

Use the template field names exactly.

Default sections:

• Risk Summary
• Allowed
• Conditional
• Blocked
• Safer Path
• Structured Risk Screen

Checklist

• Specific risks are named, not implied
• Allowed, conditional, and blocked actions are concrete
• A safer adjacent path is offered when needed
• The response avoids pretending to give formal legal advice
• Readable memo and structured JSON are both present

Common mistakes

Mistake	Fix
Vague legal warning	Name exact blocked behaviors
Treating generalized knowledge as fully blocked	Preserve lawful generalized knowledge where possible
Missing safer path	Offer a clean-room alternative, not just a stop sign

Key principles

1. Specificity matters — Name exactly what is blocked, conditional, and allowed.
2. Clean-room beats temptation — Safer adjacent paths matter more than clever justifications.
3. Generalized knowledge can remain usable — Restrict confidential assets, not lawful judgment itself.