auto-thinking-pi is a pre-release project pinned at 0.0.1. There is no stable release, no published npm artifact, and no compatibility promise. Only the current tip of main at version 0.0.1 receives best-effort security fixes.

Consumers should track main on GitHub rather than pin a version. If a security fix lands, it lands as a new commit on main and the version stays at 0.0.1 (project convention).

Please do not open a public GitHub issue for security reports.

Report vulnerabilities through GitHub's private vulnerability reporting on this repository. If that is unavailable, open a draft security advisory on the repository and include:

• a description of the vulnerability,
• steps to reproduce (a minimal prompt or input, if applicable),
• the version affected,
• any suggested remediation.

We will acknowledge your report within 7 days. Please give us a reasonable window (typically 30–90 days) to publish a fix before public disclosure.

In scope:

• The classifier (src/heuristic.ts) — e.g. catastrophic-backtracking regex, resource exhaustion on adversarial input.
• The extension (extensions/pi/index.ts) — e.g. handler crashes, unbounded memory growth in the decision ring buffer, unintended side effects on user input.
• Anything that would let a crafted prompt cause the extension to modify, drop, or exfiltrate the user's input.

Out of scope:

• The Pi coding-agent runtime itself. Report those to the Pi maintainers.
• Third-party dependencies (report upstream first, then to us if we need to update).
• Denial of service from valid but very long prompts. The classifier is O(n) in prompt length by design.