More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    mem20 audit

    1

    Audit mem20 stores for data contamination, index drift, and health metrics to ensure grounding integrity.

    $1

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    mem20 audit

    mem20 audit

    Example session with this skill installed

    Run a contamination audit on my mem20 store and check for index drift. The metrics endpoint is active at localhost:8080.

    • Read your context and instructions
    • Compiled the mem20 audit

    I have completed the audit of your mem20 store. The analysis included a scan of the grounded indexes, the BM25 corpus, and a check of the live health metrics.

    mem20 Audit Report

    MetricValueStatus
    contamination_rate0.042⚠️ VIOLATION
    Grounded Records1,240Nominal
    Simulated Records54⚠️ DETECTED
    Index DriftNone DetectedClean
    Health EndpointReachable (8080)Online

    Violation Details

    The audit detected 54 simulated records within the grounded index path /data/grounded/idx_01. These records originated from a simulated test run and have leaked into the production corpus, violating the contamination_rate = 0.0 invariant.

    Offending Sample Records:

    • sim_ref_99283: "Test vector for grounding failure..."
    • sim_ref_99284: "Synthetic prompt response for load testing..."

    Recommendation

    DO NOT TREAT AS TRUSTED. The store is currently contaminated. You must isolate the simulated material and purge it from the grounded index before using this store for high-stakes decisions or further production writes.

    Next steps

    1. Run mem20-ops recovery routines to purge the identified simulated record IDs.
    2. Review your recent bulk import logs to identify how simulated material reached grounded paths.
    3. Re-run this audit once the purge is complete to confirm a 0.0 contamination rate.

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

    About this skill

    Memory systems often suffer from silent data corruption or accidental ingestion of simulated data. This skill audits your mem20 store to verify data integrity, ensuring your grounded indexes remain free from contamination and drift.

    What it does

    • Contamination detection identifies when simulated records have leaked into grounded indexes or the BM25 corpus.
    • Index drift analysis detects discrepancies between the ledger and the indexes to prevent hidden retrieval errors.
    • Health monitoring reads live counters from the metrics endpoint to track grounding coverage.
    • Trust verification provides a go/no-go recommendation based on the contamination_rate invariant.

    How it works

    1. Initialize Audit by checking the live MCP tool list for available mem20 contamination and health methods.
    2. Execute Scans on the grounded store and BM25 corpus to locate any simulated material.
    3. Analyze Metrics via the localhost:8080/metrics endpoint or the Python API to calculate the current contamination rate.
    4. Report Findings using a standardized format that highlights violations and sample offending records.

    Frameworks & tools

    This skill works with the mem20 memory system, specifically leveraging its Python API, HTTP metrics endpoints, and MCP tool surfaces.

    Why this beats prompting it yourself

    Manual inspection of vector stores and ledgers is error-prone and misses subtle index drift. This skill enforces the mem20 zero-contamination invariant strictly, stopping operations the moment data integrity is compromised.

    Use cases

    • Verify store integrity immediately after a bulk data migration.
    • Establish a baseline audit after the initial install and first write operations.
    • Investigate unexpected recall behavior to rule out grounding coverage issues.
    • Periodically check long-running multi-user setups for ledger-index disagreement.

    Known limitations

    Requires a running mem20 process or access to the store via the Python API. It does not automatically delete or fix contaminated data.

    Requires an active mem20 process or Python API access. Will not perform automated data deletion or recovery. Localhost access is needed for metrics monitoring.

    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 8 days ago

    Requires

    mem20 or mem20-ce

    Permissions

    Read Files
    Terminal / Shell
    Write Files
    Network Access
    Environment Variables

    File Scopes

    ~/.mem20/**
    ~/.mem20-ce/**
    Listed8 days ago

    What's inside

    Frequently Asked Questions

    This skill identifies silent data contamination where simulated or synthetic records accidentally leak into your grounded indexes, and it detects index drift where your ledger and vector stores no longer align.

    This skill is designed to work with any AI agent or platform that supports MCP (Model Context Protocol) tools and has access to a running mem20 instance via Python or HTTP.

    No, this skill is a diagnostic tool; it identifies violations and provides sample offending records, allowing you to make informed decisions on how to safely clean or restore your data.

    Setup is straightforward: ensure your mem20 store is active, connect the skill to your agent's tool surface, and verify the agent has access to the metrics endpoint (defaulting to localhost:8080).

    The package includes the logic for the Contamination Detection and Index Drift modules, pre-configured MCP tool definitions, and standardized reporting templates for your audit logs.

    You should run an audit immediately after bulk migrations, following your first write operations, or anytime you notice unexpected recall behavior to rule out grounding coverage issues.

    New in LLM Ops