
Sensitive Browser
- 385 installs
- 1 repo stars
- Updated January 29, 2026
- napoleond/sensitive-browser
sensitive-browser is an agent skill that runs login, payment, and authenticated browser actions in an isolated CLI subprocess so developers can automate confidential web flows without leaking credentials into agent logs
About
sensitive-browser is a security-focused agent skill from napoleond/sensitive-browser that delegates sensitive browser work—website login, payment forms, credential entry—to a dedicated subprocess outside the core agent loop. Instead of passing passwords, session tokens, or card data through the main context, agents invoke a separate sensitive-browser CLI with isolated handling so PII and secrets stay out of shared logs, prompts, and artifacts. Developers reach for sensitive-browser when building agentic browser automation that must access authenticated or confidential pages without exposing credentials to the LLM transcript. Triggers include requests to log into websites, fill payment forms, or perform authenticated browser actions where data separation is mandatory. The pattern suits payment checkout flows, SSO logins, and any workflow where standard browser MCP tools would over-share sensitive input into agent memory.
- credential isolation
- PII redaction defaults
- scoped browser sessions
- safe screenshot policy
- audit-friendly logging
Sensitive Browser by the numbers
- 385 all-time installs (skills.sh)
- Ranked #564 of 2,203 Security skills by installs in the Skillselion catalog
- Data as of Aug 4, 2026 (Skillselion catalog sync)
npx skills add https://github.com/napoleond/sensitive-browser --skill sensitive-browserAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 385 |
|---|---|
| repo stars | ★ 1 |
| Last updated | January 29, 2026 |
| Repository | napoleond/sensitive-browser ↗ |
How do agents browse authenticated pages without leaking secrets?
Let agents browse authenticated or confidential web pages without leaking credentials, session tokens, PII, or screenshots into logs, prompts, or shared artifacts.
Who is it for?
Agent developers automating logins, payments, or confidential web forms who must keep credentials out of LLM context and shared artifacts.
Skip if: Public page scraping or testing flows with no credentials where standard browser automation tools suffice.
When should I use this skill?
User asks the agent to log in, enter payment details, or fill authenticated forms in a browser workflow.
What you get
Completed authenticated browser actions with credentials confined to an isolated subprocess, not agent logs or prompts.
- completed authenticated session
- isolated credential handling
- non-leaking automation workflow
Files
Sensitive Browser
Execute sensitive browser actions securely by delegating credential handling, payments, and form filling to a dedicated subprocess with isolated context.
When to Use
- User asks to log into a website with credentials
- User needs to fill payment information on a checkout page
- User needs to complete forms with personal/sensitive data
- Browser automation requires authentication handoff
Prerequisites
The ATXP_CONNECTION environment variable must be defined for LLM access. This provides the connection string to the ATXP platform.
If not configured, see the ATXP CLI for setup instructions.
Workflow
1. Prepare Sensitive Data File
Create a JSON file with the user's sensitive data. Ask the user for the values needed:
{
"credentials": {
"username": "user@example.com",
"password": "secret"
},
"payment": {
"cardNumber": "4111111111111111",
"expiry": "12/28",
"cvv": "123",
"billingZip": "94102"
},
"personal": {
"fullName": "Jane Doe",
"phone": "+1-555-0123",
"address": "123 Main St"
}
}Only include the fields needed for the task. Store securely and delete after use.
2. Execute the Sensitive Action
npx sensitive-browser "<task>" \
--sensitive-data ./creds.json \
--url <target-url> \
--state ./session.json \
--output ./session.jsonOptions:
| Option | Description |
|---|---|
<task> | Natural language task (e.g., "log in with credentials") |
-d, --sensitive-data <path> | Path to sensitive data JSON (required) |
-u, --url <url> | Target URL to navigate to |
-s, --state <path> | Playwright storage state to load |
-o, --output <path> | Output path for updated state |
3. Integration with agent-browser
When used with agent-browser, hand off session state:
# Save current session
agent-browser --session myapp state save ./session.json
# Hand off to sensitive-browser for login
npx sensitive-browser "log in with credentials" \
--state ./session.json \
--sensitive-data ./creds.json
# Resume agent-browser with authenticated session
agent-browser --session myapp state load ./session.jsonSecurity Notes
- Sensitive data values are never logged or displayed
- The tool shows which fields will be used, not their values
- Delete sensitive data files after use
- Consider environment variables or secret managers for production
Sensitive Data Schema
Supported field categories:
credentials: username, password, emailpayment: cardNumber, expiry, cvv, billingZip, cardholderNamepersonal: fullName, firstName, lastName, phone, address, city, state, zip, country, dateOfBirth- Custom fields: Any additional
key: valuepairs as needed
Related skills
How it compares
Pick sensitive-browser when credentials or payment data are involved; use standard browser MCP tools only for public, non-authenticated pages.
FAQ
Why use sensitive-browser instead of normal browser tools?
sensitive-browser keeps credentials, payment data, and session tokens in a dedicated subprocess so they never enter the main agent prompt, logs, or shared artifacts. Standard browser tools risk leaking secrets into LLM context during login or checkout flows.
What actions does sensitive-browser handle?
sensitive-browser handles website login, payment form filling, and other authenticated browser actions where sensitive data must stay separate. Agents delegate those steps to the CLI tool instead of typing secrets into the core conversation.