Small-Cell Suppression and Detail-Deleting Retention

    1

    A small count identifies a person: "this clinic had exactly one filed complaint in March" plus outside knowledge is a name.

    $29

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Small-Cell Suppression and Detail-Deleting Retention

    Example session with this skill installed

    We publish a public dashboard of complaints per provider (contacts, filed, founded, unfounded, pending, other). Legal flagged that small numbers could identify complainants. Add proper small-cell suppression to the JSON and CSV routes and make sure the retention job really deletes the personal details.

    • Read your context and instructions
    • Compiled the small-cell suppression

    Summary

    Suppression wired into both public routes with the outcome partition declared as a group; the audit over all 212 published rows reports no recoverable cell. Retention now clears ciphertext columns per record and keeps the counts.

    Suppression (threshold 5, marker <5)

    const GROUPS = [{ members: ['founded', 'unfounded', 'pending', 'other'], total: 'filed' }];
    rows = suppressRows(rows, { threshold: 5, groups: GROUPS });   // JSON route
    res.send(toCsv(rows, columns, { marker: '<5' }));             // CSV route
    
    providercontactsfiledfoundedunfoundedpendingother
    Clinic A4132<5<5<5<5
    Clinic B1200000
    Clinic C<5800<5<5

    Clinic A: pending was 2, so the whole outcome group is hidden although founded (10) and unfounded (10) cleared the threshold on their own; filed (32) stays visible and reveals nothing with all four addends hidden. Clinic C: the two zeros stay; pending + other = 8 is one equation with two unknowns.

    auditSuppression over every row: 0 violations. Added a test that fails if either route stops calling the suppressor or either query drops other.

    Methodology note published: "Counts greater than zero and smaller than 5 are withheld and shown as <5. Zero is always shown. Where several figures add up to a published total (founded + unfounded + pending + other = filed), the whole set is withheld when any one of them is small..."

    Retention

    • Details are encrypted per record (HKDF(master, recordId) -> AES-256-GCM); the worker clears contact_ciphertext, narrative_ciphertext, transcript_ciphertext, deletes the encrypted files, expires download tokens, sets sensitive_deleted_at, keeps provider, category, rule code, event month, filing state, outcome, timing, language.
    • Windows: completed 72h, delivered-not-filed 72h, authorized-not-confirmed 7d, incomplete 72h. One audit event per record with no personal data.
    • Backups: retention set to the window plus the backup cycle, or the deletion promise is false.

    Needs sign-off: the threshold and the windows are policy; record the approval on the go-live checklist.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    A small count identifies a person: "this clinic had exactly one filed complaint in March" plus outside knowledge is a name. And the details behind the counts outlive their purpose unless deletion is designed in. This skill is both halves.

    Suppression (suppress.mjs)

    Counts above zero and below the threshold (5 by default, the common public-health baseline) are withheld; zero is always shown; fields that partition a published total (founded + unfounded + pending + other = filed) are hidden together when any member is small, zeros inside stay visible, and a small total hides itself and the group. auditSuppression proves no hidden cell is recoverable by subtraction; suppressRows for tables; toCsv prints a marker, never an empty cell or the word null; methodologyNote writes the paragraph to publish beside the numbers.

    Retention (field-crypto.mjs)

    Per-record keys from HKDF-SHA-256 of a master key and the record id, AES-256-GCM envelopes with a random nonce, cross-record decryption fails, SHA-256 of the plaintext for integrity, keyed HMAC fingerprints so identities match without being stored. retentionPlan gives the deletion time by state (completed, delivered-not-filed, authorized-not-confirmed, incomplete), the columns to clear, the analytic columns to keep, and the audit event.

    Reference

    The seven suppression rules with the reasoning, what may stay in analytics and what never may, the retention worker's steps, evidence-intake validation order, and why backup retention must match the deletion promise.

    Requirements

    7 offline tests. Node 20+, no packages, no network. Engineering controls, not legal advice.

    Handles counts and single-level groups; no nested hierarchy support. Pure logic only; user must execute SQL/file deletes. Thresholds and windows require local policy approval.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    Requires

    Node.js 20+

    Permissions

    Terminal / Shell

    Shell runs node for the tests. The library is pure functions; it reads and writes no files, makes no network calls and reads no environment variables. The master key for the retention half is passed in by your code from wherever you keep secrets.

    Listed19 days ago

    What's inside

    Frequently Asked Questions

    This skill provides the logic to hide low-frequency data points (typically <5) and uses grouped partitioning to ensure hidden values cannot be calculated by subtracting known sub-groups from a total.

    The skill is a Node.js library compatible with Node 20+ and can be integrated into any AI agent or backend service running a Javascript/TypeScript environment. It relies on native modules, meaning no external NPM dependencies or network calls are required for it to function.

    Yes, the library includes a per-record encryption utility using AES-256-GCM and HKDF keys, allowing you to securely store PII while keeping non-identifying counts accessible for analytics.

    You receive the complete Node.js library, including the suppression engine, the auditing tool for compliance checks, cryptographic utilities for field-level encryption, and implementation templates for JSON/CSV outputs.

    While the skill provides the logic for suppression and identifies which records to purge, you are responsible for executing the actual SQL 'DELETE' or 'UPDATE' commands and managing file system deletions within your specific database environment.

    New in LLM Ops