What a security reviewer can rely on today. Vulnerability reporting is in SECURITY.md.
| Control | Where |
|---|---|
| Publishing without stored secrets | crates.io, PyPI, npm, RubyGems and NuGet use trusted publishing from the release workflow (Releasing) |
| Build provenance | npm packages publish with --provenance; PyPI uploads carry attestations from trusted publishing |
| Tag-only releases | The release workflow runs on a tag push; nothing is published by hand |
| Pinned actions | Third-party GitHub Actions in the release workflow are pinned to a commit SHA |
| Dependency updates | Dependabot opens weekly pull requests for Cargo and GitHub Actions |
| Memory safety | The core forbids unsafe (unsafe_code = "forbid"); checks that need it live in an unpublished crate (a real fork, a counting allocator), and the runtime’s worker and queue tests run under miri in CI |
| Secret scanning | .gitguardian.yaml configures scanning; .gitignore excludes key and environment files |
| No required runtime dependencies in the host language | Python and npm packages carry the compiled core instead of pulling a dependency tree |
| Data minimisation | Prompt and completion content is off by default; secrets and PII are redacted in Rust before any transport sees a record |
| Tamper evidence | Optional hash-chained, signable logs, checked with logquill verify |
The privacy, redaction and tamper-evidence features are technical controls that commonly support compliance work (for example access, retention and audit requirements). Compliance is a property of a whole system and process, not of a library. LogQuill does not make an application HIPAA-, PCI- or GDPR-compliant.
The 3.0 line is MIT licensed (see LICENSE). Earlier Python 2.0.0 stays FSL-1.1-MIT and all other earlier releases stay MIT, as published.
CHANGELOG.md.npm audit signatures in a project that depends on the package.logquill verify <file>.