# Privacy and sensitive input

Every API call is remote processing at `https://api.tools.11dz.cn`. Content sent to it leaves the current machine. Do not infer that a transformation is local merely because the Skill is installed locally.

Before calling:

1. Confirm that the user explicitly selected 11DZ remote processing.
2. Identify secrets, credentials, personal data, customer data, or confidential business fields.
3. Prefer a synthetic sample or redact confidential values while retaining the required structure.
4. Never send private keys, seed phrases, passwords, cookies, session keys, active production JWTs, or production credentials.
5. If potentially confidential content is still required, explain the remote destination and obtain explicit user direction before proceeding.
6. Keep request JSON in stdin or a protected file instead of a shell argument.
7. Never echo a complete sensitive request body in diagnostics or chat output.

## Bundled client guard

The client scans parsed field names and common secret shapes such as private-key headers, access-key formats, credential assignments, and compact JWTs. It also treats every `/v1/developer/jwt/decode` call as sensitive.

Without `--confirm-sensitive`, a guarded request exits before network access with `CLIENT_SENSITIVE_INPUT_CONFIRMATION_REQUIRED`. Do not add the flag automatically. First redact the input or obtain explicit user authorization for that specific remote transmission.

The scan is a safety backstop, not a secret detector with complete coverage. A request that passes the scan can still be confidential; apply human and task context before every call.

The API does not require or accept an API key through this client. Never add an `Authorization` header.

Use the public browser tools at `https://tools.11dz.cn` when local-only processing is required; browser tools remain local by default.