This repository is intended to publish verification and evaluation tooling, not operational execution infrastructure.
The stable public package may include read-only provider observations, privacy-minimal evidence models, deterministic local benchmarks, local sandbox tooling, offline trace adapters, and documentation needed to understand those interfaces.
The stable public package must not contain live provider-mutation controllers, live model/provider execution bridges, credentials, tokens, secret-discovery logic, production target configuration, customer/personal traces, private prompts, or deployment/security topology.
See docs/PUBLIC_SURFACE.md for the publication rules enforced by the release tests.
Do not post credentials, private identifiers, exploit details, customer data, or other sensitive material in a public issue or pull request.
Use GitHub's private vulnerability-reporting feature when it is available for this repository. If it is not available, use an already-established private channel with the repository owner and keep the initial report minimal until a private route is confirmed.
Assume anything committed to this public repository can be copied permanently. Secrets must be rotated rather than merely deleted if they are ever exposed.
The authenticated read-only verifier accepts credentials from caller-owned code at runtime; credentials are not discovered from environment variables or local secret stores by the stable package and are excluded from default public evidence.
Security reports are especially useful for:
- credential or caller-controlled identifier leakage;
- fail-open verification behavior;
- unsafe path or symlink handling;
- remote-state confusion or repository-identity mistakes;
- release artifacts that expose private inputs;
- accidental addition of provider write capability, live execution bridges or secret discovery to the stable public package.
This policy does not claim that a verifier match proves causality, user authorization, intent, permanence, or production safety.