Works with the AI Agents you already use
Redaction Proof
It commits the original to a SHA-256 fingerprint, removes the spans, and writes a proof that anyone can verify against the redacted file without ever seeing the original.
$12
Redaction Proof
Example session with this skill installed
I need to send incident report IR-2291 (original.txt) to the client, but it's got customer details in it. Strip the personal data and give me something they can check. Just emails for now is fine.
- Read your context and instructions
- Compiled the redaction proof
I ran redaction-proof redact on original.txt with --preset email. Verdict: FAIL (exit code 1): the emails were removed, but the residual scan found five more items of personal data still in the redacted text.
{
"kind": "redaction-proof",
"verdict": "FAIL",
"why": "5 item(s) that look like personal data are still in the redacted text",
"spanCount": 2,
"linesHidden": [2, 8],
"linesRevealed": 7,
"root": "85c0b76161e30a6f040309a6656e6b65aad83466855584b995f66591ac73bb71",
"spanSeal": "3e0ac02be126828dc880581e211fb6bdca0c962c25e47b1589b327e25ad9800a",
"residual": [
{ "label": "phone", "line": 2, "text": "07700 900123" },
{ "label": "card", "line": 3, "text": "4111 1111 1111 1111" },
{ "label": "uk-postcode", "line": 4, "text": "LS1 4AB" },
{ "label": "uk-nino", "line": 5, "text": "QQ 12 34 56 C" },
{ "label": "phone", "line": 8, "text": "+44 20 7946 0958" }
],
"proofSha256": "fa100bef257c066c451ff5a8d2201c193034831600cfd5910073de50168c636f",
"warnings": [ "no --key: a hidden line that is short or guessable can be brute-forced from its hash. Use --key for personal data and keep the key private" ]
}
What happened
- Both email addresses were redacted and sealed.
- A mobile number, a card number, a postcode, an NI number and a landline are still in the text.
- No key was used, so a short hidden line could be guessed from its hash.
I'd rerun with --preset email,phone,uk-nino,card,uk-postcode, a --pattern for the reporter's name and street, and --key-file. That run came back PASS with 9 spans redacted and a clean residual scan, and the client'sverifyreturned FAITHFUL against the root.
Connects securely to your tools. The creator never sees your data.
About this skill
Agent Trust: redact it, and prove you only removed what you said
When you send someone a redacted document, they have to take your word that you only took out the personal data. This skill gives them a way to check. It commits the original to a SHA-256 fingerprint, removes the spans, and writes a proof that anyone can verify against the redacted file without ever seeing the original.
What it does
- Redacts by preset (email, phone, UK National Insurance number, card number with Luhn check, UK postcode, IPv4), by your own regex, or by exact character spans. Each removal becomes a numbered
[[Rn]]marker. - Commits the original as a Merkle root over its lines. Untouched lines are revealed in the proof; lines with a redaction are committed by hash only.
- Seals the removed spans and their positions in a SHA-256 hash chain.
- Scans the redacted output for anything that still looks like personal data and fails if it finds some.
- Verifies: the recipient checks the redacted file against the proof (and the root, if you sent it ahead). FAITHFUL, BREACHED or MALFORMED, with reasons.
When an agent should use it
- Sharing incident reports, support logs, call transcripts or case notes outside your team
- Answering a data request where the other side needs to trust the redaction
- Checking a redacted file someone else sent you
What it outputs
redacted.txt, proof.json, and JSON with PASS/FAIL, span count, hidden and revealed lines, the root, the span seal, residual findings and SHA-256 hashes. verify-original lets the holder of the original re-check everything, including the seal. Exit 0 = PASS/FAITHFUL, 1 = FAIL/BREACHED.
Limitations
- Presets are pattern matches. Names, street addresses and free-text identifiers need a
--patternor spans. A PASS means the scan found nothing it recognises, not that the text is anonymous. - The proof reveals the position and length of each removed span.
- Without a key, a hidden line that is short or predictable can be guessed from its hash. Use
--key-filefor personal data and keep the key private. - Plain text only; convert PDFs or Word files to text first.
Open source, packaged skill
Original work by Simon Gant. The core logic is public under the MIT licence (kar-redactseal and kar-veil). This listing is the packaged, tested agent skill: SKILL.md workflow, CLI, presets, residual scan, keyed salting, a SHA-256 span seal and docs. Requires Node.js 18+, zero dependencies, no network, no API keys.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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 today
- Software
Node.js 18+
Permissions
Runs one local Node.js script (`node scripts/redaction-proof.js`) through the agent's shell tool. Reads only the files you pass it (original, spans, key file, proof, redacted text). Writes `redacted.txt` and `proof.json` only into the `--out-dir` you name. No network access, no environment variables read. Pure compute.