agentleFS
lawalOyinlolalawalOyinlola/appsec-protocols/skills/legal-compliance/SKILL.md

legal-compliance

Skill1 starsChanged 50 days agoMITView on GitHub

Baseline legal and regulatory controls for any app used by real people — privacy policy, terms, AI disclosure, arbitration, auto-renewal, UGC/DMCA, app-store privacy labels, and jurisdiction scoping. Load BEFORE writing or editing legal pages, launching, opening signups, adding an AI feature, adding uploads, adding subscriptions, or submitting to an app store or platform review (Meta, Apple, Google, Stripe). Triggers on "privacy policy", "terms of service", "ToS", "arbitration", "DMCA", "UGC", "AI disclosure", "FTC", "GDPR", "CCPA", "COPPA", "privacy nutrition label", "data safety", "auto-renew", "click to cancel", "cookie banner", "am I going to get sued", "is this legal", "compliance", plus any task touching legal pages, subscriptions, uploads, or personal data.

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 10 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 10

    Distinctive

    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:

    Also: no dead href="#" links in the footer.

    line 60
  2. Distinctive:

    Every service that receives personal data: email, SMS/WhatsApp, analytics, maps, hosting, payments, error tracking, AI providers.

    line 75
  3. Very specific:

    Verify: write the four answers down.

    line 116
  4. Distinctive:

    Verify: every capability claim on the landing page maps to shipped code.

    line 129
  5. Distinctive:

    Verify: if added, it must be conspicuous and mutual.

    line 137
  6. Distinctive:

    If billing recurs: disclose price, interval and renewal *before* purchase, and make cancellation reachable without a support ticket.

    line 143
  7. Distinctive:

    Verify: check current thresholds; do not assume.

    line 160
  8. Distinctive:

    If AI touches user data: name the provider as a processor (Group A.4), say whether inputs train third-party models, and disclose AI …

    line 205
  9. Distinctive:

    Verify: check the provider's data-retention terms and whether a zero-retention or no-training option exists.

    line 207
  10. Distinctive:

    If under-13 (US) or under-16 (parts of EU) users are plausible, COPPA/GDPR-K obligations attach and they are heavy.

    line 216
  11. Distinctive:

    Never handle raw card data — use a processor's hosted flow so PCI scope stays with them.

    line 222
  12. Distinctive:

    Never report "legally compliant" — that is not a thing you can determine.

    line 233

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “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

Legal Compliance

Baseline legal controls for software that real people use. Companion to security-protocols: that skill keeps data from leaking, this one keeps the product from being unlawful or deceptive. Every control has a verification step — a control is not "done" until it has been proven by reading the live page, the code, or the filing.

How to use this skill

  • Writing or editing legal pages → work Group A, then whichever of C–F the jurisdiction gate (Group B) actually opens.
  • Pre-launch / pre-platform-review → work all groups in order, write results to tasks/legal-audit.md (one line per control: PASS / FAIL / N/A + why).
  • "Not applicable" is a valid and common answer — most controls here do not apply to most apps — but it must be justified in one line. Silence is not.

Two rules that override everything else in this skill

1. You are not a lawyer, and neither is the user's source. This skill scopes and prioritises; it does not give legal advice. When exposure is real and material — a regulator, a class of consumers, employee surveillance, children's data, health or financial data — say plainly that this needs a qualified lawyer in the relevant jurisdiction, and keep going with what you can do. Never let "get a lawyer" become a reason to deliver nothing.

2. Verify the claim before acting on it. Legal claims reach users through viral videos, Twitter threads and vendor marketing, and they are frequently wrong, stale, or jurisdiction-confused in ways that produce real wasted work. Before implementing any specific claim, check three things:

  • Is the rule current? Rules get struck down. The FTC's "click-to-cancel" Negative Option Rule, for instance, was vacated by the Eighth Circuit in 2025 — citing it as binding federal law would be wrong. Search for the rule's current status.
  • Does it reach this user? Nearly every alarming figure in circulation is US federal or state law. A company incorporated outside the US, serving non-US customers, is often simply outside that reach. Establish jurisdiction (Group B) first.
  • Is the penalty figure real? Statutory maxima are per-violation ceilings under specific statutes or consent orders, adjusted annually — not the automatic price of a missing sentence. Never repeat a dollar figure you have not sourced.

Tell the user plainly when a claim they brought does not apply to them, and why. Agreeing that everything applies is not caution — it is wasted work and false comfort, and it crowds out the risks that are real for them.


Group A — Foundations (every app with real users)

1. Privacy policy exists and is publicly reachable

Required by essentially every platform (Meta, Apple, Google, Stripe) and most privacy law.

  • Verify: fetch it signed-out and assert a 200, not a redirect: curl -s -o /dev/null -w "%{http_code} %{redirect_url}\n" https://<domain>/privacy
  • Common failure: the page exists but sits behind auth middleware, so reviewers and crawlers get bounced to /login. Check the route-protection allowlist explicitly.
  • Also: no dead href="#" links in the footer.

2. Terms of service exists and is publicly reachable

Terms are what let you suspend abusive accounts and cap liability.

  • Verify: same check as control 1, against the terms route: curl -s -o /dev/null -w "%{http_code} %{redirect_url}\n" https://<domain>/terms
  • Same common failure: reachable while signed in, bounced to /login when signed out.

3. State what data is collected, specifically

Enumerate real categories from the actual schema — not boilerplate. Read the data model before writing this section.

  • Verify: diff the policy's list against the database schema and any analytics/SDK calls. Anything collected and unlisted is a defect.

4. Name third-party processors

Every service that receives personal data: email, SMS/WhatsApp, analytics, maps, hosting, payments, error tracking, AI providers.

  • Verify: cross-check against the CSP connect-src, the dependency list, and env vars for third-party API keys. Those three catch what memory misses.

5. Deletion, export, and correction path

Say how a user asks, and what happens on account closure. Retention claims must match reality.

  • Verify: find the code that actually deletes. If a cleanup job only prunes some tables, the policy must not claim more. A retention promise the code does not keep is a false statement in a privacy policy — the worst kind of defect here, because it is written down and dated.

6. Working contact channel

A real, monitored address — not a role alias that forwards nowhere.

  • Verify: send a message to the published address from an unrelated account and confirm it arrives. An address that bounces is worse than none: it is a published promise you are visibly failing, and it is the channel a regulator or platform reviewer will try first.

7. Identify the legal entity accurately

Legal name, registered address, and registration number where applicable — matching the incorporation documents exactly.

  • Verify: compare against the certificate of incorporation, character for character. Trading names, abbreviations, and punctuation variants all cause platform rejections.
  • Watch for: registered address on the certificate differing from the operating address. Pick one, use it consistently everywhere, and expect to explain the other.

Group B — Jurisdiction gate (decide before applying C–F)

8. Establish reach before applying any rule

Answer these four, in the project's docs, before implementing anything from Groups C–F:

QuestionWhy it decides things
Where is the entity incorporated?Sets default governing law and regulator
Where do users actually live?US → FTC/state law. EU/UK → GDPR. Both → both
B2C or B2B?Class actions and consumer-protection rules mostly target B2C
How is it distributed?App Store/Play impose their own regime; web-only does not
  • Verify: write the four answers down. If the answer is "no US users", most US-specific controls below are N/A and should be recorded as such rather than implemented.
  • Governing law + venue clause in the Terms is the cheapest control in this skill and does much of the work that panicked founders attribute to arbitration clauses.

Group C — US-facing (only if serving US users)

9. Do not make deceptive claims, including about AI

The FTC's actual concern is deception, not the absence of a disclosure sentence. Claiming capability the product lacks ("AI-powered" when it is if/else, "bank-level security") is the violation.

  • Verify: every capability claim on the landing page maps to shipped code.
  • If the product genuinely uses AI: say so where a user would reasonably want to know — especially if AI output affects decisions about them, or if they might mistake it for a human. This is cheap and worth doing regardless of jurisdiction.

10. Arbitration and class-action waiver

Enforceable in the US and genuinely limits class exposure for B2C products. Much less relevant B2B, and unenforceable against consumers in much of the EU/UK.

  • Verify: if added, it must be conspicuous and mutual. A buried clause is struck.
  • Judgment: for a non-US B2B product, a governing-law and exclusive-jurisdiction clause usually matters more. Do not bolt a US arbitration clause onto a product with no US users and call it protection.

11. Subscription auto-renewal and cancellation

If billing recurs: disclose price, interval and renewal before purchase, and make cancellation reachable without a support ticket.

  • Verify: click through the real cancel flow. Count the steps against signup.
  • Status check: the FTC's click-to-cancel rule was vacated in 2025, but state laws (notably California's ARL) still bind, as do Apple/Google/Stripe rules. Check current status rather than citing the federal rule.

12. DMCA designated agent — only if hosting user uploads

Safe harbour under 17 U.S.C. §512 requires registering an agent with the US Copyright Office (small fee, renewable) and a takedown process in the Terms.

  • Verify: does the app actually accept user-uploaded content? Grep for upload handlers and file inputs. No uploads → N/A. This is the control most often cargo-culted onto apps that have no UGC at all.

13. State privacy laws (CCPA/CPRA and successors)

Trigger on thresholds (revenue, volume of consumers, data sales). Below threshold → record N/A with the reason.

  • Verify: check current thresholds; do not assume. If in scope: "Do Not Sell/Share" link, opt-out honouring, and a rights-request path.

Group D — EU/UK (only if serving EU/UK users)

14. GDPR essentials

Lawful basis per purpose; controller vs processor roles stated; processor agreements in place; data-subject rights honoured within statutory deadlines; breach notification path; international transfer safeguards.

  • Verify: the policy names a lawful basis, not just a purpose. For B2B, state clearly which data the customer controls and which you control — this determines who answers a data-subject request.

15. Cookies and tracking

Consent required before setting non-essential cookies. Strictly-necessary cookies (session, CSRF) do not need consent and should be described as such.

  • Verify: load the site fresh and inspect what is set before any interaction.

Group E — App store distribution (only if shipping a native app)

16. Apple Privacy Nutrition Label / Google Play Data Safety

Required for App Store and Play distribution. Must list every collected data type, including what SDKs collect on your behalf — analytics, ads, crash reporting.

  • Verify: enumerate SDKs, then reconcile against the declared label. An undeclared SDK is a rejection and a misrepresentation.
  • Web apps and PWAs distributed outside the stores → N/A. A manifest.webmanifest is not store distribution.

Group F — Product-specific high risk

17. Location, biometric, and health data

Location history is sensitive: it reveals home, routine, associations. Biometric data carries statutory damages under laws such as Illinois BIPA.

  • Verify: state retention plainly, keep tenants isolated, and never repurpose it.
  • Employee/driver tracking is its own risk, separate from customer privacy: workers usually must be informed, and in some jurisdictions must consent. Put the obligation on the customer explicitly in the Terms, and give them the wording to pass on.

18. AI features

If AI touches user data: name the provider as a processor (Group A.4), say whether inputs train third-party models, and disclose AI involvement where a user would care.

  • Verify: check the provider's data-retention terms and whether a zero-retention or no-training option exists. If user content is sent to a model, the policy must say so.
  • Disclosure is the legal half; the technical half is security-protocols 29–30 (prompt injection, and per-user cost caps). Neither substitutes for the other.
  • If AI output could affect safety — health, legal, financial, crisis or self-harm contexts — there must be a defined safe response and a human escalation path. This is a product requirement, not a legal footnote.

19. Children

If under-13 (US) or under-16 (parts of EU) users are plausible, COPPA/GDPR-K obligations attach and they are heavy.

  • Verify: state a minimum age in the Terms and mean it. "Not intended for children" in a policy while the product courts them is exactly the deception regulators pursue.

20. Payments

Never handle raw card data — use a processor's hosted flow so PCI scope stays with them.

  • Verify: confirm no PAN reaches your servers or logs. State in the policy that the processor handles payment details.
  • security-protocols 42 is the matching technical control — server-side pricing, and the signed webhook as the source of truth for "paid". A checkout the client can re-price is a consumer-protection problem as well as a security one.

Reporting

Never report "legally compliant" — that is not a thing you can determine. Report per control, with evidence:

A.1 Privacy policy reachable   PASS  curl → 200, no redirect, signed out
A.5 Retention matches code     FAIL  policy claims 90-day deletion; cleanup job
                                     only prunes stale invites, positions kept
C.12 DMCA agent                N/A   no user uploads (no upload handlers in src)

Then state, in one sentence each: what is genuinely exposed, what is N/A and why, and what needs a real lawyer.

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.