regulator-eyes
GitHub模拟监管视角（如FTC）审计营销文案，识别明示/暗示声明及遗漏风险。输出证据需求清单、质询函草稿及整改建议，帮助用户规避合规隐患。
Trigger Scenarios
Install
npx skills add mohitagw15856/pm-claude-skills --skill regulator-eyes -g -y
SKILL.md
Frontmatter
{
"name": "regulator-eyes",
"description": "Read your marketing claims, landing page, or ad copy the way a consumer-protection investigator would (FTC\/ASA framing) and draft the inquiry letter they could send. Use when asked to check my marketing claims, read this like a regulator, audit my landing page for claim risk, or is this ad compliant. Produces a claim inventory with substantiation demands, the inquiry letter, and a fix-or-drop debrief per claim."
}
Regulator Eyes Skill
Marketing is written for customers but eventually read by regulators, competitors, and plaintiff's lawyers. This skill performs that hostile reading now: every claim inventoried, the substantiation each would require, and the inquiry letter that arrives when someone files a complaint. (Environmental claims have a dedicated sibling: greenwashing-self-audit.)
What This Skill Produces
- Claim inventory — every express and implied claim, including ones made by images, testimonials, and omission
- Substantiation demands — what evidence a regulator would require per claim, and whether the user has it
- The inquiry letter — the civil investigative demand / information request they could receive
- Fix-or-drop debrief — per claim: keep with evidence, reword, add disclosure, or drop
Required Inputs
Ask for these if not provided:
- The marketing material — landing page text, ad copy, emails, app store listing (paste it)
- What evidence exists — studies, data, guarantees infrastructure (or "none yet" — that's an answer)
- Jurisdiction/vertical (optional) — default to US FTC framing; flag if health, finance, or children's products (higher bar)
Framework: How an Investigator Reads
| Pass | Looking for |
|---|---|
| 1. Express claims | Direct statements: "fastest", "clinically proven", "saves 40%", "#1" |
| 2. Implied claims | What a reasonable consumer takes away — before/afters, testimonials as typical results, comparison imagery |
| 3. Material omissions | Conditions, fees, auto-renewals, "results not typical" realities left unsaid |
| 4. Format traps | Fake countdown timers, dark-pattern cancellation, undisclosed endorsements/affiliates |
Risk scale: 🔴 enforcement-grade (deceptive on its face or unsubstantiated health/money claim) · 🟡 challengeable (defensible only with evidence the user must produce) · 🟢 puffery (opinion no reasonable consumer takes literally — "the best coffee in town").
Judge claims by the net impression on a reasonable consumer, not the writer's intent — that is the actual legal standard's shape.
Output Format
Regulatory Reading: [Asset] — [date]
Simulation — a plausible adversarial reading, not a prediction or legal advice.
Claim Inventory
| # | Claim (verbatim) | Type (express/implied/omission) | Substantiation required | User has it? | Risk |
|---|
The Inquiry Letter
[A formal information request citing the specific claims, demanding the substantiation, with a response deadline — the document that starts a very bad quarter.]
Debrief — out of character
| # | Verdict | New wording or required disclosure |
|---|---|---|
| [keep / reword / disclose / drop for every 🔴 and 🟡] |
Confirm anything load-bearing with an advertising-law attorney — standards vary by jurisdiction and vertical.
Quality Checks
- Implied claims and omissions are inventoried, not just literal sentences
- Every 🔴 names the specific missing substantiation, not "needs evidence"
- Puffery is honestly rated 🟢 — inflating everything to red destroys the signal
- The letter cites the user's actual claims verbatim
- Every red/yellow claim gets a concrete verdict with replacement wording where kept
Anti-Patterns
- Do not grade intent — grade the net impression on a reasonable consumer
- Do not invent claims the material doesn't make; the inventory quotes the source
- Do not offer "add an asterisk" as a fix for a deceptive net impression — disclosures cure omissions, not lies
- Do not treat testimonials as safe because they're "just customers talking" — typicality is the user's problem
- Do not stay in character in the debrief
Version History
- 54fad50 Current 2026-07-19 13:39
