Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Czech Document Anonymization

    1

    Anonymise Czech documents before they reach an AI model, then verify with checksums that no personal data slipped through.

    Free

    2 usesSecurity scanned
    Czech Document Anonymization

    Czech Document Anonymization

    Example session with this skill installed

    Anonymise this court ruling before I send it to an AI model for a summary, and check that nothing personal is left.

    "Rozsudek ve věci žalobkyně Jany Novákové, rodné číslo …, bytem Masarykova 12, Brno, tel. +420 777 123 456, e-mail jana.novakova@example.cz, proti žalované společnosti Stavby Morava s.r.o., IČO: 25596641, IBAN CZ65 0800 0000 1920 0014 5399. Okresní soud v Brně, sp. zn. 12 C 345/2026."

    • Read your context and instructions
    • Compiled the czech document anonymization

    Anonymised (placeholder mode, consistent across the document)

    Rozsudek ve věci žalobkyně

    OSOBA1 OSOBA2, rodné číslo

    RC1, bytem

    ULICE1 ENTITA1,

    MESTO1, tel.

    TELEFON1, e-mail

    EMAIL1, proti žalované společnosti

    FIRMA1, IČO:

    ICO1, IBAN

    IBAN1.

    INSTITUCE1, sp. zn.

    SPZN1.

    18 replacements: 8 by the structured pre-pass (phone, e-mail, IČO, IBAN, account, case number, court), the rest by the NER model.

    Leak check (scripts/pii_leak_check.py, offline)

    TextPossible leaks
    Original6 — IČO, IBAN, bank account, e-mail, phone, date of birth
    Anonymised0

    Checksums (mod-11, mod-97, Luhn) keep false alarms low; excerpts are masked, e.g. IČ•••••••••41.

    Next step: a quick human read for rare surnames and addresses, then send the anonymised text to the model. Keep the mapping file private — it re-identifies people.

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

    About this skill

    Protect sensitive data before sending it to LLMs or third parties. This skill handles the redaction of Czech personal identifiers, names, and locations using a deterministic placeholder system. It solves the risk of leaking PII (Personally Identifiable Information) in documents like court rulings, medical reports, and bank statements while maintaining the document's internal logic and readability.

    What it does

    • PII Redaction replaces names, addresses, and organizations with consistent placeholders like OSOBA1 or MESTO1.
    • Checksum Verification runs secondary offline checks for birth numbers (rodné číslo), IČO, IBAN, and credit card numbers using mathematical validation.
    • Multi-language Support identifies entities in Czech plus 30 other languages using LINDAT NLP services.
    • Context Preservation keeps clinical codes (ICD-10) and statute citations intact to ensure the document remains useful for analysis.
    • Morphological Analysis extracts lemmas and parts of speech to help understand document structure without exposing identities.

    How it works

    1. Selection of the anonymization mode, choosing between the default API or the Zero-egress offline mode for highly sensitive data.
    2. Execution of the anonymize command to generate deterministic placeholders across the entire text.
    3. Verification using the bundled pii_leak_check.py script to find surviving identifiers via regex and checksums.
    4. Validation through extract_entities to confirm no remaining personal names or specific locations exist.

    Frameworks & tools

    This skill utilizes the anonymize-mcp server, LINDAT NLP tools (NameTag, UDPipe), and a specialized Python verification script.

    Why this beats prompting it yourself

    Standard LLM prompts often hallucinate PII or miss identifiers hidden in complex Czech grammar. This skill uses deterministic NLP models and mathematical checksums to ensure that birth numbers and bank accounts are actually validated, not just guessed at.

    Use cases

    • Redacting medical reports before generating an AI summary.
    • Preparing court rulings for public disclosure by removing names and dates.
    • Scrubbing bank statements of account numbers and IBANs for financial analysis.
    • Cleaning HR documents before sharing them with external recruiters.

    Known limitations

    Non-commercial use only due to CC BY-NC-SA model licenses. Offline mode disables translation and readability tools to prevent data egress. Requires local installation of anonymize-mcp.

    Non-commercial use only (licence of the underlying NameTag/UDPipe models). Default mode sends text to the public LINDAT API in Prague; use ANONYMIZE_MCP_LOCAL=1 to stay offline, which disables translation, spell check and readability. Anonymisation lowers risk but is not a legal guarantee — rare surnames and addresses still need a human read.

    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

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Anonymize MCP
    Python 3.9+
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Terminal runs the bundled offline leak checker (scripts/pii_leak_check.py). Read/Write Files: the document you anonymise and the anonymised copy. Network Access is used only in the default API mode (text sent to LINDAT/ÚFAL in Prague); with ANONYMIZE_MCP_LOCAL=1 nothing leaves your machine.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    It uses LINDAT's NameTag and MasKIT models, trained on Czech, plus a pre-pass of 80+ patterns for structured data (birth numbers, IČO, IBAN, phones, case numbers). The bundled checker then validates leftovers by checksum (mod-11, mod-97, Luhn), so you verify the result instead of trusting a single pass. It lowers risk; a human should still read anything released under GDPR.

    Yes. With ANONYMIZE_MCP_LOCAL=1, anonymisation and entity extraction run fully offline with a local model (one-time ~31 MB download, SHA-256 checked). Tools that would send text out (translation, spell check, readability) are refused in that mode, and the bundled leak checker is offline by design.

    Yes, the skill uses a deterministic placeholder system (e.g., OSOBA1, OSOBA2). If the same name appears multiple times throughout a document, it will be replaced by the same consistent tag, allowing an LLM to follow the narrative or logic without knowing the actual identity.

    The skill folder: SKILL.md (the workflow for the agent), README.md (install and set-up), the offline leak checker scripts/pii_leak_check.py and its tests. The anonymize-mcp server itself is free (MIT) and installed separately with pip install anonymize-mcp. The skill is free; the underlying language models are licensed for non-commercial use only.

    Any agent that reads SKILL.md and can use an MCP server: Claude Code, Codex, Cursor, Gemini CLI, Hermes Agent and others. The leak checker runs anywhere with Python 3.9+.

    While the primary NLP engine is optimized for Czech, the LINDAT integration supports entity recognition for 30 additional languages, making it suitable for multilingual documents containing Czech, English, German, and more.

    New in Legal & Compliance