Only the latest 0.1.x release receives security fixes. Upgrade before filing a report if you are on 0.0.x.

Please do not file a public GitHub issue for a suspected vulnerability.

Use one of the following, in order of preference:

1. GitHub private vulnerability reporting: open the "Security" tab on lognd/typani, then "Report a vulnerability". This creates a private advisory thread that only the maintainer and you can see.
2. Email logan@logand.app if you cannot use GitHub's reporting flow.

• A minimal reproduction (code snippet or a small script).
• The typani version and backend (python -c "import typani; print(typani.__version__, typani.backend_name())").
• Python version and operating system.
• What you expected versus what happened, and why you believe it is a security issue rather than a correctness bug.

The maintainer aims to acknowledge a report within 7 days. There is no bounty program. A fix timeline depends on severity and will be discussed with you in the private advisory thread.

typani is an in-memory value-type library with no network access, no file I/O beyond what a caller explicitly does with the values it hands back, and no code execution of untrusted input in its normal (non-lint) code paths. Given that shape, most legitimate reports will fall into one of these areas:

• The native extension. The optional typani-core crate (crates/typani-core) is a PyO3 extension. unsafe_code = "forbid" is set at the crate level, so memory-safety issues would most likely come from the PyO3 Python/Rust boundary itself rather than hand-written unsafe code. Reports here should include whether the issue reproduces under the pure-Python backend (TYPANI_PURE=1) as well.
• The backend selector. typani._impl chooses between the native and pure-Python backends at import time based on typani_core being importable and version-matched to typani itself, and falls back to pure-Python on any mismatch or forced TYPANI_PURE=1. A report claiming the selector picks an incompatible or stale backend should include the exact versions of both packages involved.
• Pickling. Ok/Err/Some/Nothing are picklable. Unpickling untrusted data is never safe in Python regardless of typani; this is not typani-specific and will be closed as out of scope unless typani's own __reduce__/__setstate__ implementation does something beyond the standard unsafe-by-design pickle contract (for example, executing code during unpickling that a plain attribute-restoring __setstate__ would not).
• typani.lint. python -m typani.lint parses source files with ast.parse only; it never executes, imports, or evaluates the code it scans. A report claiming otherwise (arbitrary code execution while linting a file) is treated as high severity.

If you are unsure whether something is in scope, report it anyway and let the maintainer make the call -- a false positive costs little, a missed report costs a lot.