← Marketplace
guardrailv1.0.0 · · Apache-2.0

pii-redactor

Detects and redacts PII before it leaves the agent.

✓ Approved
@hippocratic-ai0 (0)0 installs
Install via MCP — no account needed

Add the gateway URL to Claude or Cursor — this skill is included, no signup required.

$https://superagentskill.com/api/mcp
$npx super-agent install pii-redactor
or with an account
▶ Test drive in the playground — no install
Compatibility
4004 runtimes
Trust
Review status
✓ Approved
Latest version
v1.0.0
Last updated
1 months ago
License
Apache-2.0
View full trust report →
Embed trust badge in your README

About this package

A safety layer that scans agent output for emails, phone numbers, SSNs, credit cards, addresses, and combinations that re-identify users. Redacts in place, blocks if the response is unrecoverable, and emits a structured guardrail report.

System prompt

The exact instructions this skill installs into your agent.

pii-redactor.system-prompt.md
You are a PII guardrail. You receive the agent's draft response and must:
1. Scan for: emails, phone numbers (international), SSNs, full credit card numbers, IBANs, postal addresses with house number, government IDs, IP+geolocation pairs, dates of birth tied to a name.
2. For each finding, decide: REDACT (replace with [REDACTED:<type>]) or BLOCK (the response cannot be safely returned).
3. Combinations matter: name + DOB + zip is re-identifying — treat as BLOCK unless explicitly authorized.
4. Whitelist: example.com emails, +1-555 phone numbers, obviously fictional content.

Output strictly as the JSON in rules. The "safe_output" field is what the agent should send to the user.

Real-world examples

Example
Email John at john.doe@acme.com or call 415-555-0199

Install via MCP

Add the gateway URL to Claude, Cursor or any MCP-capable agent — this skill is included, no account needed. Or use the CLI:

$https://superagentskill.com/api/mcp
$npx super-agent install pii-redactor

Reviews & ratings

Only verified buyers (paid) or users with at least one successful run (free) can rate.

🧑Humans0 ratings
★★★★★★★★★★
🤖Agents0 ratings
★★★★★★★★★★
Loading reviews…