Use GitHub's private vulnerability reporting — the Security tab, then Report a vulnerability. That keeps the report out of the public issue tracker until there is something to say about it.

If that is not available, email ehsun.ca@gmail.com. Please do not open a public issue for a vulnerability first.

Expect an acknowledgement within a week. This is a 0.x project maintained by one person in their own time, and pretending to a response time it cannot meet would be worse than saying so.

Fixes land on main and go out in the next 0.8.x. While the version stays below 1.0, a security fix may arrive alongside behavioural changes rather than on its own — pinning an exact version and never updating is the wrong shape of caution here, because there is no branch to backport to and there will not be one until the interface stops moving.

Worth being concrete, because "a physics simulator" sounds like it has no attack surface and that is not quite true.

A .maiman project and a .pdk kit are JSON. Loading one may only name components that are already in the registry; it can never import a dotted path. So opening a file someone sent you is not equivalent to running their code, and "type": "os.system" is refused by name like any other unregistered component — there are tests for exactly that in tests/test_project.py and tests/test_pdk.py.

What a hostile file can do is describe a very expensive simulation. See below.

maiman serve starts an HTTP server whose /api/run accepts a graph and executes it. That is the point of it, and it is why the socket binds to 127.0.0.1 unless a host is given explicitly. Passing --host something else prints a warning, deliberately.

Do not expose it to a network you do not control. There is no authentication, and there is not meant to be: it is a local tool whose front end happens to be a browser page.

A graph is arbitrary work. A large enough window will occupy a core for a long time and allocate accordingly, which is a denial of service against the machine running it whether the request came from a person or a file. The server refuses a run whose window exceeds MAX_SAMPLES (4,194,304 samples) before it starts, rather than discovering the problem while allocating.

That limit is a guard rail, not a sandbox. Long sweeps and deep spans can still be slow by design.

src/maiman/studio/index.html is a single file with no build step and no runtime dependency on a CDN. It talks to the local server and to nothing else.

• A wrong physical result. That is a bug, and an important one — open a public issue with the reference it should match. See CONTRIBUTING.md.
• A slow simulation. Unless it is disproportionate to the window, in which case it is a performance bug.
• The server executing a graph you sent it. That is its documented purpose on loopback.

The runtime dependency is NumPy, and that is deliberate: the fewer things in the tree, the fewer things to audit. FFTW is deliberately absent — it is GPL-2.0-or-later, and linking it would change this project's licence rather than its security, but the effect on the dependency surface is the same. Development additionally uses pytest, mypy, ruff, build, hatchling and PyYAML.

PyYAML is dev-only and worth one line about why it is not a runtime dependency. maiman.netlist reads the YAML netlists a layout tool writes, and reading one must not be able to execute anything — so the reader works over a parsed mapping, handles JSON with the standard library, and calls yaml.safe_load rather than yaml.load on the path where a YAML parser is present at all. A netlist, a .maiman project and a .pdk are all documents, and opening one somebody sent you is not equivalent to running their code.