According to 1M AI News monitoring, hours after the new Claude Code pet feature/buddy system went live, the developer community found a “re-roll” method from leaked source code; open-source tools and tutorials have already spread across platforms such as Linux.do, V2EX, GitHub, and others.
Currently, there are at least two re-roll paths. One targets the fixed salt value in the pet generation algorithm. This value is hard-coded with 15 characters in Claude Code’s compiled output; by replacing with the same length, you can change the random seed. After brute-force enumeration, you can find the value corresponding to the target pet and write it in. The other leverages differences in the authentication path: when subscribed users log in normally, the server sends accountUuid as the pet seed, and it can’t be tampered with. But when authentication is done via environment variables, that value isn’t written into the local configuration; the system then falls back to reading the userID field, which can be freely edited.
The method that changes the salt is the most fragile—next time Claude Code updates, it will be overwritten. Changing userID also has an expiration window for subscribed users; Anthropic can simply add logic to write the environment-variable-path accountUuid to close the loophole. API users (authenticated with their own API Key) naturally don’t have an accountUuid, so they can directly modify userID, giving them a relatively longer window; but Anthropic can also block it by generating a stable identifier from the API Key hash. Skeleton data isn’t stored locally; it’s computed in real time at each startup, so any server-side patch can take effect immediately.
Additionally, userID isn’t only used for the pet system. Leaked source code shows that telemetry reporting and A/B test bucket assignments also read this value; modifying it could cause experimental features to malfunction or lead to a break in usage data continuity.