agentleFS
BenPremBenPrem/business-brain/.claude/skills/regulated-copy-compliance/SKILL.md

regulated-copy-compliance

Skill1 starsChanged 3 months agoMITView on GitHub

Pre-publish compliance scan for any client asset touching a regulated topic — financing, health claims, legal services, insurance. Maintains a per-client ruleset (trigger terms, prohibited phrases, required disclosures), scans files/folders/URLs, and outputs PASS/FAIL with file:line citations. Triggers - "compliance check", "compliance scan", "can we say this", any client asset that mentions a regulated topic. NOT general site QA — use site-qa-checklist.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every client with regulated exposure gets clients/<slug>/compliance/ruleset.md.

    line 16
  2. Very specific:

    Write to clients/<slug>/compliance/scans/YYYY-MM-DD-<asset-slug>.md (mkdir -p the scans/ dir first) and summarize in chat.

    line 94
  3. Distinctive:

    Default to the safe side and route the question to the client's compliance contact via the founder — never resolve regulatory ambiguity by …

    line 107
  4. Distinctive:

    Commit the ruleset, then wire the hook gate.

    line 125
  5. Distinctive:

    After any scan that blocks or clears a deliverable: log an activity note on the client's workspace record (date, asset, verdict, blocking …

    line 135

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Regulated Copy Compliance

Pre-publish scan for client assets that touch a regulated topic. Fires on ANY asset — website, landing page, ad, email, SMS, social post, graphic, video script, flyer, sign, radio/TV spot, in-home sales material — that mentions or implies the regulated subject. Enforcement is the agent's job: the client will not remember these rules mid-campaign, and the founder should not have to. The scan catches it before anyone sees it.

The per-client ruleset (the binding document)

Every client with regulated exposure gets clients/<slug>/compliance/ruleset.md. Read it fresh at the start of every scan — never work from memory of it. No ruleset yet → build one first (see "Building a ruleset" below), get the founder's sign-off, then scan.

A ruleset carries six sections:

  1. Governing sources — the actual program guide, ad policy, or regulation the rules derive from (link or archived copy). Every rule below cites one.
  2. Trigger terms — phrases/patterns that, once used, escalate disclosure requirements (e.g. in consumer-credit advertising: any rate, term length, or payment amount).
  3. Prohibited phrases — the hard DON'T list. Any hit is a FAIL regardless of context.
  4. Required elements when the regulated program/partner IS named — verbatim disclosures, marks, logos, conspicuousness rules. Paraphrase = FAIL.
  5. Standing policy — the client's chosen posture (usually: generic language only in public media; the branded path reserved for approved sales material).
  6. Compliance contact — the named human whose WRITTEN approval clears branded assets.

Inputs

A file, a folder, or a URL. For URLs: curl -sL <url> -o <scratch>/scan-<slug>.html and scan the fetched file — the deployed page is ground truth; prefer it over local source when both exist. For folders, scan *.html *.md *.txt *.json recursively. For video or graphics, extract the transcript and on-screen text first (video-analyzer), then scan those.

Scan procedure (all greps -inE, case-insensitive, line-numbered)

Step 0 — topic detection. Grep for the ruleset's topic-detection pattern (e.g. for financing: financ|apr|loan|credit|per month|/mo|payment plan|monthly payment plus the partner's name). No hits → report "no regulated content detected — scan not applicable" and stop.

Step 1 — trigger terms. Grep every pattern in ruleset §2. Any hit means the matching required-disclosure obligation from §4 now applies to this asset — record which.

Step 2 — prohibited phrases. Grep every entry in ruleset §3. Each hit = FAIL with the exact offending string and file:line.

Step 3 — interactive claim tools. Calculators, estimators, sliders, quizzes that output a regulated figure (a payment amount, a health outcome, a settlement estimate): grep calculat|estimat(e|or) near the topic, plus <input type="range"> / JS that computes the figure. Most program guides ban these outright — default FAIL unless the ruleset explicitly permits them.

Step 4 — required elements (only when the program/partner is named). Check every §4 item: correct name with ® mark where required, verbatim disclosure fingerprint (grep a distinctive substring of the exact required text — a paraphrase won't match, which is the point), "subject to approval"-type qualifiers wherever terms appear, required logos (check text, alt attributes, and image filenames), localized disclosures on translated pages. Conspicuousness rules (disclosure near the claim, readable size, contrasting, not hidden in a modal) can't be grepped — render/screenshot the page and check by eye.

Worked example — financing-partner branding

The most common ruleset shape for home-services clients with a third-party financing partner:

  • Standing policy: the lending partner is NEVER named in public-facing media. Public assets use generic language only — "Financing available", "Ask us about financing options." A public asset that names the lender is a policy FAIL even if its disclosures are technically complete.
  • Trigger terms (from consumer-credit advertising rules): any specific rate or APR, any term length or number of payments, any monthly-payment dollar amount, down-payment claims, deferred-interest headlines ("no interest until…"). One hit requires the full plan disclosure on the SAME asset, using partner-supplied plan numbers — hand-filled numbers = FAIL.
  • Prohibited phrases: "no credit check", "guaranteed approval", "pre-approved", "same as cash", "easy financing", "as low as", standalone "0%" / "no interest" without its qualifying condition, false-urgency promo deadlines that don't really expire.
  • Calculators: payment estimators on the client's site are banned — any hit = FAIL.
  • Branded path (approved sales material, or if the client reverses the standing policy): partner named correctly with its mark, "subject to credit approval" wherever plans appear, the program's verbatim disclosure block, required lender logos, and the compliance contact's written approval on file. Until that approval exists, the only correct status is "drafted, pending compliance approval" — never "done" or "live".

Adapt the same skeleton for health claims (substantiation, before/after imagery, FTC guidance), legal services (state bar advertising rules: "specialist" claims, outcome guarantees, required disclaimers), and insurance (state-level ad rules, licensing lines).

Output — the report

Write to clients/<slug>/compliance/scans/YYYY-MM-DD-<asset-slug>.md (mkdir -p the scans/ dir first) and summarize in chat. One row per ruleset checklist item → PASS / FAIL / N/A, every FAIL with file:line and the exact offending string. Then an overall verdict:

  • PASS (generic path): topic mentioned in generic terms only, no partner named, zero trigger terms, no banned tools — compliant by construction under the standing policy.
  • FAIL: any Step 1–3 hit or missing Step 4 element. List the minimum edits that reach the generic-safe path.
  • BLOCKED ON APPROVAL: asset names the partner/program and is otherwise clean — status stays "drafted, pending compliance approval" until written sign-off exists.

When unsure whether a phrase is a trigger term: treat it as non-compliant and say so. Default to the safe side and route the question to the client's compliance contact via the founder — never resolve regulatory ambiguity by yourself.

Hook wiring (make the gate mechanical)

tools/hooks/deploy-guard.sh carries a commented client-specific compliance-gate example: it blocks any deploy for a named client unless a scan report newer than 24h exists in clients/<slug>/compliance/scans/. When a client's ruleset goes live, uncomment and adapt that block — a gate that depends on remembering to run it is not a gate.

Building a ruleset for a new client/topic

  1. Get the governing documents (partner program guide, platform ad policy, statute/agency guidance) — from the client, the partner rep, or primary-source research. Never draft rules from general knowledge alone.
  2. Extract trigger terms, prohibitions, and required disclosures into the six-section format, each rule citing its source section.
  3. Have the founder confirm the standing policy and the compliance contact with the client.
  4. Commit the ruleset, then wire the hook gate.

Invocation from other skills

website-builder, site-qa-checklist, social and email skills, proposal-generator — any skill producing client-facing media for a client that has a ruleset MUST run this scan whenever the regulated topic appears in the asset. Don't wait to be asked; the gate fires automatically.

Write-back

After any scan that blocks or clears a deliverable: log an activity note on the client's workspace record (date, asset, verdict, blocking items) and update the matching <TASK SYSTEM> task with the verdict.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.