cctrace intercepts real, credentialed traffic between Claude Code and Anthropic. We take that seriously. This document explains what the tool does with sensitive data and how to report a problem.

Every captured request/response pair passes through a single redaction step (src/redact.ts) before it reaches any sink — the .jsonl log, the self-contained .html snapshot, or the live WebSocket:

• Headers — authorization, x-api-key, cookie, and similar are masked to a first-10/last-4 preview.
• Bodies — credential fields (access_token, refresh_token, client_secret, code, api_key, …) are masked in JSON and form-encoded bodies. Conversation content in /v1/messages is left intact.
• URLs — credential-bearing query params (e.g. OAuth ?code=) are masked.

The CA and leaf private keys are generated locally under .cache/mitm/ with 0600 permissions and never leave your machine. The .cctrace/ output directory is gitignored by default.

• A trace is still a record of your real session. Review it before sharing. Redaction targets known credential shapes; it is not a guarantee that a novel secret embedded in a message body will be caught.
• The MITM CA, while trusted only by the Claude process cctrace launches, exists on disk while cctrace runs. Keep .cache/mitm/ private.

Please do not open a public issue for security problems.

Use GitHub's private vulnerability reporting on this repository: Security → Report a vulnerability (https://github.com/thevibeworks/cctrace/security/advisories/new).

Include what you observed, how to reproduce it, and the impact. We aim to acknowledge within a few days. Credential-leak-to-disk reports are treated as the highest priority.

cctrace is pre-1.0; security fixes land on the latest release. Please run the newest tag before reporting.