Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    ToxicJoin Compositional SQL Privacy Review

    2

    Review AI-generated SQL for compositional privacy and re-identification risks using DataHub governance, field classifications, lineage analysis, and verified subject-threshold checks.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    ToxicJoin Compositional SQL Privacy Review

    ToxicJoin Compositional SQL Privacy Review

    Example session with this skill installed

    Review the following AI-generated SQL query for compositional privacy risks.

    Use the following synthetic DataHub metadata fixture:

    • subject_token: STABLE_PSEUDONYM
    • postal_prefix: QUASI_IDENTIFIER
    • age_band: QUASI_IDENTIFIER
    • support_category: SENSITIVE_ATTRIBUTE

    The query is

    SELECT
        e.subject_token,
        e.postal_prefix,
        e.age_band,
        e.support_category
    FROM demo_events AS e;
    

    Determine whether the combination creates a privacy disclosure risk.

    Explain the field classifications, query output grain, relevant risk pattern, and recommended remediation.

    Do not invent missing metadata, claim a real-world re-identification event, or authorize execution. Clearly distinguish this synthetic example from a live DataHub assessment.

    • Read your context and instructions
    • Compiled the toxicjoin compositional sql

    ToxicJoin — Compositional Privacy Review

    1. Executive Summary

    Assessment: BLOCK_LIKE

    The submitted SQL exposes a stable pseudonym alongside multiple quasi-identifiers and a sensitive attribute at individual-row granularity.

    This combination creates a compositional privacy disclosure risk.

    The assessment is based on the synthetic governance metadata supplied in the request.

    2. SQL Structure Analysis

    The query selects four fields from demo_events.

    FieldGovernance ClassificationAssessment
    subject_tokenSTABLE_PSEUDONYMStable subject linkage
    postal_prefixQUASI_IDENTIFIERGeographic linkage
    age_bandQUASI_IDENTIFIERDemographic linkage
    support_categorySENSITIVE_ATTRIBUTESensitive disclosure

    Output grain: Individual rows.

    The statement contains no GROUP BY or aggregation that would transform the output into a grouped analytical result.

    3. Identified Risk

    Finding: TJ-PRIV-001

    Risk pattern

    STABLE_PSEUDONYM + MULTIPLE QUASI_IDENTIFIERS + SENSITIVE_ATTRIBUTE

    The selected attributes create a combination that can increase the potential for subject linkage and disclosure.

    Although no direct identifier appears in the projection, the stable pseudonym and quasi-identifiers provide a potentially linkable subject context.

    The available evidence establishes a risky disclosure condition. It does not establish that an actual person has been re-identified.

    4. Recommended Remediation

    Recommended actions

    1. Review whether the stable pseudonym is necessary for the analytical purpose.
    2. Evaluate whether geographic and demographic fields can be reduced or coarsened under the organization's approved policy.
    3. Consider a supported grouped analytical design where appropriate.
    4. Verify all transformed fields against their DataHub classifications and upstream lineage.
    5. Reparse and reassess the complete modified SQL before execution.

    No arbitrary privacy threshold is proposed because the configured minimum and verified subject key were not supplied.

    5. Evidence Limitations

    This demonstration uses synthetic metadata supplied in the request.

    No live DataHub lookup, warehouse query, or external re-identification test was performed.

    A production assessment requires actual governed metadata, lineage evidence, and a compatible SQL AST parser.

    6. Final Decision

    ControlResult
    Compositional risk observedYES
    Sensitive attribute exposedYES
    Individual-row outputYES
    Live DataHub verificationNOT PERFORMED
    Execution authorizationNOT GRANTED
    ToxicJoin runtime decisionNOT EVALUATED

    Final Assessment: BLOCK_LIKE

    The query should be referred to the deterministic ToxicJoin enforcement engine and the responsible privacy policy owner before execution or release.

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

    About this skill

    ToxicJoin Compositional SQL Privacy Review

    Overview

    AI data agents can create serious privacy risks even when individual database fields appear harmless on their own.

    ToxicJoin Compositional SQL Privacy Review helps identify these risks by examining how identifiers, pseudonyms, quasi-identifiers, and sensitive attributes combine across SQL queries.

    It uses governed DataHub metadata, schema classifications, and data lineage to produce evidence-based privacy assessments.

    What This Skill Does

    • Compositional Privacy Analysis: Detects risky combinations of fields that may expose sensitive information.
    • DataHub Governance: Reviews field classifications, metadata, and glossary terms.
    • Data Lineage Analysis: Examines upstream dependencies and data relationships.
    • SQL Structural Analysis: Reviews projections, joins, CTEs, filters, grouping, aggregates, and HAVING conditions.
    • Subject Threshold Verification: Checks whether COUNT(DISTINCT subject_key) uses the correct governed privacy subject and a properly structured threshold.
    • Evidence-Based Reporting: Produces structured findings, missing evidence, risk explanations, and remediation recommendations.

    Why It Matters

    Traditional SQL reviews often focus on individual columns or obvious SQL injection patterns.

    ToxicJoin focuses on compositional privacy risks: combinations of otherwise acceptable fields that may enable subject identification or expose sensitive attributes.

    For example, a stable pseudonym combined with geographic and demographic attributes may create a disclosure risk even when no direct identifier is selected.

    Review Outcomes

    The skill uses four assessment categories:

    • BLOCK_LIKE
    • REWRITE_CANDIDATE
    • NO_TRIGGER_OBSERVED
    • INSUFFICIENT_EVIDENCE

    Every assessment includes its supporting evidence and limitations.

    Requirements

    • Read-only DataHub MCP access or an equivalent metadata interface.
    • A SQL AST parser compatible with the submitted SQL dialect.
    • An organization-approved field classification map.
    • A configured minimum distinct-subject threshold.

    Important Security Boundary

    This skill is an advisory review assistant. It does not execute SQL, modify DataHub metadata, authorize data release, or replace the deterministic ToxicJoin enforcement engine.

    It never issues an authoritative ALLOW decision.

    Missing or ambiguous evidence results in INSUFFICIENT_EVIDENCE rather than an assumed safe outcome.

    Project

    Developed from the ToxicJoin open-source project:

    https://github.com/Z3X-1337/toxicjoin

    License: Apache-2.0

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Requires DataHub metadata access and a SQL AST parser. It cannot authorize query execution or guarantee absolute anonymity. Does not support dynamic SQL or unresolved bindings.

    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 buying a skill.

    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 17 days ago

    Software
    DataHub MCP
    SQLGlot
    orother SQL AST parser

    DataHub needs read-only metadata and lineage access. Also needs an approved data classification map and a configured minimum distinct-subject threshold.

    Permissions

    Network Access

    Allowed Hosts

    datahub.com

    Read-only DataHub metadata and lineage access is required for evidence-based governance reviews. The skill does not execute SQL, modify warehouse data, write DataHub metadata, access credentials, or authorize data release. If required tools or governance evidence are unavailable, the skill returns INSUFFICIENT_EVIDENCE.

    Listed17 days ago

    What's inside

    Frequently Asked Questions

    No. It reviews SQL structure and governance metadata without executing queries or modifying warehouse data.

    Yes, for complete governance and lineage verification. Without access to the required metadata, it reports insufficient evidence.

    No. It produces advisory findings. Authoritative enforcement requires the deterministic ToxicJoin runtime.

    No. It identifies supported compositional privacy risks but does not provide a universal anonymity or regulatory compliance guarantee.

    AI agents that support the universal SKILL.md format and can access the required external tools.

    New in Agents & Orchestration