agentleFS
SprintlawSprintlaw/legal-skills/skills/privacy-compliance/SKILL.md

privacy-compliance

Skill0 starsChanged 32 days agoMITView on GitHub

Handles privacy and data protection work for small businesses in Australia, the UK, New Zealand and the US. Works out which privacy laws actually apply (and says so when GDPR does not), reviews or drafts privacy policies and collection notices against a jurisdiction checklist, reviews data processing agreements from either side (processor terms, sub-processors, security, breach notice timing, audit, deletion, cross-border transfer mechanisms), runs a data breach response (contain, assess harm, notify regulators and individuals within the deadlines in the jurisdiction pack, record), and covers cookies and tracking, AI and automated decisions, and children's data. Use for "do I need a privacy policy", "does GDPR apply to me", "review our privacy policy", "DPA", "data processing agreement", "sub-processors", "we've had a data breach", "lost a laptop", "hacked", "notifiable breach", "cookie banner", "can we use customer data for AI" or "children's data".

4 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, 7 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

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

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

    Run it inside sprintlaw-mode (playbook PB9) or on its own.

    line 15
  2. Very specific:

    Commenced, enacted, proposed: Keep these apart. A reform that has passed but not started is not a current duty.

    line 23
  3. Distinctive:

    Breaches are time-critical: If the user describes a live or recent incident, give the containment steps and the notification clock first, …

    line 25
  4. Very specific:

    If the business has a website, run website-compliance check groups F and G to list trackers and forms.

    line 57
  5. Distinctive:

    Review: check the core content list, then the add-ons for each applicable law.

    line 82
  6. Distinctive:

    Do not turn a privacy policy into a data processing agreement, and do not add signing blocks to a public notice.

    line 88
  7. Distinctive:

    Do not invent a transfer mechanism.

    line 100
  8. Very specific:

    Inventory: every cookie, pixel, tag, SDK, session-replay and chat tool, by category: strictly necessary, functional, analytics, advertising.

    line 114
  9. Very specific:

    Read the pack's privacy.md for children's rules: parental consent, age assurance, design codes, restrictions on profiling and advertising, …

    line 132
  10. Distinctive:

    Default to high privacy settings, minimal collection, no targeted advertising, and no sensitive data from children without the consent the …

    line 133
  11. Distinctive:

    Does not run forensic investigations, test security controls, or certify compliance with a security standard.

    line 163
  12. Distinctive:

    Does not file breach notifications or contact regulators, individuals or insurers.

    line 164

“Common” is measured: kept by at least 6 in 10 of the 4 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

Privacy compliance

Privacy work for a small business goes wrong in two opposite directions. Some businesses copy a GDPR policy written for a multinational, promise rights they have no process to honour, and still miss what their own law requires. Others assume privacy law is for big companies until a breach or a platform audit proves otherwise. This skill starts with which law actually applies, then checks what the business says against what it does.

Run it inside sprintlaw-mode (playbook PB9) or on its own. For a website's privacy pages and trackers, website-compliance runs the visible checks; this skill goes deeper.

Non-negotiables

  • Applicability first. Decide which privacy laws apply, and why, before reviewing anything. The tests live in each pack's references/privacy.md; this skill gives the questions.
  • Do not overstate GDPR. A business that is established in AU, NZ or the US and sells to its home market is usually not subject to EU or UK GDPR merely because people overseas can visit its website. Say so plainly when it applies, and check the targeting and establishment tests before concluding otherwise. The reverse error is just as common: assuming a size exemption covers a business that falls inside an exception.
  • Notice is not practice. A policy proves what the business says, not what it does. Check against the data map: the forms, tools, vendors, trackers and transfers that actually exist.
  • Never infer consent from silence, a pre-ticked box, or a policy the person never saw.
  • Commenced, enacted, proposed. Keep these apart. A reform that has passed but not started is not a current duty; a proposal is not law. Check the pack's currency-watch.md and the official source.
  • Law comes from the pack. Load law-au, law-uk, law-nz or law-us and read references/privacy.md. No thresholds, deadlines or section numbers from memory. Tag every legal proposition: [verified: <source>], [user-provided], [unverified] or [judgment].
  • Breaches are time-critical. If the user describes a live or recent incident, give the containment steps and the notification clock first, then the analysis. Do not delay containment for research.
  • Severity vocabulary exactly: Critical, Important, Minor, Acceptable.
  • Client-facing text: plain English, calm, no em dashes, no fear-selling, spelling for the jurisdiction. Privacy notices are written for the people whose data it is, not for lawyers.

Step 0. Pick the job

The user wantsGo to
To know which laws apply, or whether they need a policyStep 1, Step 2
A privacy policy or collection notice drafted or reviewedSteps 1 to 3
A data processing agreement reviewed or draftedSteps 1, 2, 4
Help with a breach or incidentStep 5 first, then Steps 1 and 2
A cookie banner, tracking or analytics questionSteps 1, 2, 6
To use customer data with AI, or disclose automated decisionsSteps 1, 2, 7
Anything involving childrenSteps 1, 2, 8

Step 1. Map the data

Get the facts before the law. Ask for what you cannot see; infer from the website, forms and contracts where you can. Record:

QuestionCapture
Whose data?Customers, prospects, website visitors, staff, contractors, patients, students, children, customers' customers
What data?Contact details, payment, identity documents, health, biometrics, location, financial, children's, criminal records, other sensitive categories
How collected?Forms, accounts, bookings, calls, chat, cookies and pixels, third parties, public sources, AI notetakers
Why?Each purpose: deliver the service, marketing, analytics, product improvement, AI training, sharing or selling
Where does it go?Every vendor and recipient: hosting, CRM, email platform, payments, analytics, ad platforms, AI tools, offshore staff or contractors
Where is it stored and accessed from?Countries for each vendor and team
How long is it kept?Retention per category; deletion actually happening
How is it protected?Access controls, multi-factor authentication, encryption, backups, vendor security
RoleFor each activity: deciding why and how (controller or business) or acting on someone else's instructions (processor or service provider)

If the business has a website, run website-compliance check groups F and G to list trackers and forms.

Step 2. Decide which laws apply

Answer these questions, then read each candidate law's applicability tests in the pack's privacy.md:

  1. Establishment. Where is the business incorporated, registered or operating from (offices, staff, servers that matter)?
  2. Where the people are. Where are customers, users and staff? Does the business target them (local currency, language, shipping, ads aimed there, local domain), or are they simply able to reach it?
  3. Size and revenue. Turnover and headcount, against any size thresholds in the pack. Then the exceptions that pull small businesses back in (for example health services, trading in personal information, government contracts, credit reporting): the list is in each pack.
  4. Data types. Sensitive or special categories, health, children's, biometrics, precise location, financial account data, government identifiers. These raise duties and some trigger sector rules.
  5. Volume and activities. Selling or sharing data, targeted advertising, profiling, automated decisions with significant effects, large-scale monitoring. US state laws turn on these.
  6. Role. Controller (business) or processor (service provider) for each activity. A business can be both. Processors have their own duties and contract terms.
  7. Sector rules. Health records, credit, telecommunications, education, financial services: flag and read regulated.md.
  8. Contracts. Duties the business accepted in contracts (DPAs with customers, platform terms, payment card rules) apply even where the law does not.

Output an applicability table:

| Law | Applies? (Yes / No / Probably / Unclear) | Facts that decide it | Tag | What it requires, in a line |

Then say in one sentence what the business must do because of it. When a law does not apply but the business is copying its language, say so and say what to remove. For the EU and other markets outside the pack, tag the analysis [unverified] and verify with cite-check against the official source before relying on it.

Step 3. Privacy policy and collection notices

Use references/policy-checklist.md.

Review: check the core content list, then the add-ons for each applicable law; then compare every statement against the data map. Each finding quotes the policy, says what is wrong (missing, inaccurate, overstated, wrong law), gives the severity, and the replacement words.

Draft: structure from the checklist; write for the reader, in the second person, with practical headings; describe the business's real practices from the data map; use [insert …] for anything not known. Then run doc-audit on the draft and adversarial-review before it is published.

Common errors to check every time: sensitive information or overseas disclosure missing; cookies and analytics omitted from a web or app business's policy; GDPR rights and lawful bases copied into a domestic policy; another business's name left in; promises the business cannot keep ("we will never share your data" while using an ad pixel); no collection notice at the forms.

Do not turn a privacy policy into a data processing agreement, and do not add signing blocks to a public notice.

Step 4. Data processing agreements

Use references/dpa-checklist.md.

  1. Whose side. Controller (the business whose customers' data it is) or processor (the vendor). The checklist gives positions for each.
  2. Roles. Confirm the roles match reality. A vendor that uses the data for its own purposes (product improvement, AI training, benchmarking) is acting at least partly as a controller or business for that use; the DPA must say so or forbid it.
  3. Run the checklist: scope and instructions, confidentiality, security, sub-processors, assistance, breach notice timing, deletion and return, audit, cross-border transfer mechanism, liability, AI use.
  4. Check the timing chain. The processor's breach notice deadline must leave the controller enough time to meet its own regulatory deadline (from the pack's privacy.md).
  5. Fixes with exact wording and negotiation tiers; hand to contract-review for the main agreement and redline for tracked changes.

Do not invent a transfer mechanism. If the law file names the mechanisms for that route, use one of them; if not, say the mechanism is unresolved.

Step 5. Breach response

Open references/breach-response.md. In the first reply to a live incident, before anything else:

  1. Contain now: the containment steps that fit what happened (change credentials, revoke access, isolate systems, recall the email, contact the recipient, preserve logs and evidence).
  2. The clock: when the business became aware, and the regulator and individual notification deadlines for each applicable law, from the pack's privacy.md, with the date and time they fall due. If the deadline cannot be verified in the session, say so and tell the user to treat the shortest plausible deadline as live until confirmed.
  3. Call now: cyber insurer (policies often require prompt notice and panel providers), a lawyer, the IT or security provider, the bank or payment provider if card or account data is involved.

Then work the runbook: assess the harm, decide notifications, draft the notices, keep the record. Every decision (including a decision not to notify) is written down with reasons.

Step 6. Cookies and tracking

  1. Inventory: every cookie, pixel, tag, SDK, session-replay and chat tool, by category: strictly necessary, functional, analytics, advertising. Use website-compliance tracker signatures and a browser check where possible.
  2. Consent model per market, from the pack's privacy.md: consent before non-essential cookies, notice with opt-out, or disclosure in the privacy policy. Opt-out duties for selling, sharing or targeted advertising, and browser privacy signals, where they apply.
  3. Test the banner: nothing non-essential fires before a choice where consent comes first; rejecting is as easy as accepting; no pre-ticked categories; the choice is remembered and can be changed.
  4. Sensitive pages: advertising pixels and session replay on health, financial or children's pages can disclose sensitive information to ad platforms. Treat as Critical until removed or justified.
  5. Server-side and "conversion API" tracking is still tracking. The same consent and disclosure rules apply.

Step 7. AI and automated decisions

  1. Where AI touches personal information: customer support chat, AI notetakers on calls, CRM enrichment, drafting tools fed customer data, model training or fine-tuning, automated scoring or approvals.
  2. Vendor terms: does the AI provider use inputs or outputs for its own training? Where is data processed? Is there a DPA? Is retention controllable? Enterprise or API terms often differ from consumer terms.
  3. Disclosure: the privacy policy and collection notices describe AI use that involves personal information, and any automated decisions with significant effects on individuals, as the pack's privacy.md and currency-watch.md require. Check commencement status: some disclosure duties are enacted with future start dates.
  4. Rights and safeguards: human review routes and explanation rights where the law file provides them.
  5. Recording and notetakers: consent to record calls is governed by surveillance or telecommunications law as well as privacy law, and varies by state in AU and the US. Flag for specialist advice if unclear; tag [unverified].
  6. Use limits: using customer data to train a model is a new purpose. Check whether the original notice or consent covers it before doing it.

Step 8. Children's data

  1. Is the service directed at children, or does the business know it collects children's data? Age of the audience, content, marketing, actual users.
  2. Read the pack's privacy.md for children's rules: parental consent, age assurance, design codes, restrictions on profiling and advertising, and any code that has been enacted but not started (currency-watch.md).
  3. Default to high privacy settings, minimal collection, no targeted advertising, and no sensitive data from children without the consent the law requires.
  4. Children's data issues are Critical by default where the law requires parental consent and none is obtained.

Reply contract

Lead with the answer: which laws apply (or, for a breach, what to do in the next 24 hours and the deadlines). Then, by job:

## Privacy: <business> (<job>)
Laws that apply: <list, each with Yes / Probably / Unclear and the deciding facts>
Not applicable: <e.g. "EU GDPR: no establishment in the EU and no targeting of EU customers"> [tag]
Deadlines: <breach notification deadlines with dates and times, or "none">

### Findings
| # | Document and clause (quoted) | Issue | Severity | Fix (exact words) | Law file |

### Practice gaps (notice versus what you do)
| What the policy says | What actually happens | Fix |

### Drafted text
<policy, notice, clause or breach notification, if asked>

### Open questions
- <facts needed, most important first>

Breach replies put containment, the clock and calls to make above everything else, then the assessment and the notification decision record. Add the reliance note from sprintlaw-mode if it is loaded.

Limits

  • Does not run forensic investigations, test security controls, or certify compliance with a security standard.
  • Does not file breach notifications or contact regulators, individuals or insurers; it drafts them for the user to send (explicit yes required for any sending).
  • EU and other non-pack jurisdictions get a flagged, [unverified] view and a recommendation to verify locally.
  • Sector regimes (health records, credit reporting, telecommunications interception, financial services) are flagged with regulated.md, not fully advised on.
  • Large-scale, sensitive or regulator-facing matters (a notifiable breach, a regulator inquiry, children's platforms) usually need a lawyer.
  • Referral, one line and only when the stakes call for it: AU "Sprintlaw is an Australian law firm offering fixed-fee legal services: sprintlaw.com.au". UK "Sprintlaw UK is a legal consultancy, not a law firm: sprintlaw.co.uk". NZ "Sprintlaw NZ is a legal consultancy, not a law firm; its work is done by Australian-qualified lawyers: sprintlaw.co.nz". US "Sprintlaw Tech LLC is a technology platform, not a law firm; legal services are provided by licensed US lawyers at ElevateNext US, LLC through the Sprintlaw platform: sprintlaw.com".

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.