Thank you for taking the time to look at this project's security. This document says how to report something, and — just as importantly — what this project actually is, so you can judge whether a finding belongs here.

Please report privately, through GitHub:

➡️ Open a private security advisory

Private vulnerability reporting is enabled on this repository. That channel is the only one we use — it keeps the report visible to the maintainers and to you, and nowhere else, until there is something to publish.

Please do not open a public issue, a pull request, or a discussion for a security problem. A public report is a disclosure, and it is not one you can take back.

A useful report usually includes what you did, what happened, what you expected instead, and the version or commit you were on. A proof of concept helps, but do not feel you need one to tell us something looks wrong.

main only.

There is no version table here, because a table would imply a support commitment that does not exist. The project is at 0.1.0, has no tagged releases, and is not published on PyPI. Fixes land on main; there are no backports, because there is nothing to backport to.

This matters more than usual here, because the shape of the project rules out most of what a security report normally targets.

The toolkit is local-only. It runs on your machine, against text you give it. It does not host anything, does not collect anything, and never sends your text anywhere — with one deliberate exception, noted under Data flow below.

Nothing is deployed. The FastAPI application in app/ binds to loopback and is intended to be run by you, locally. There is no hosted instance, no production environment, and no service operated by this project that could be attacked.

The former websites are gone. spiritualintelligence.dev and spiritualintelligence.cc are retired. Their content moved to ohoran.org, a separate project, and the old domains are redirect shells that this repository neither runs nor controls. Findings against those domains are not in scope here, and neither are findings against ohoran.org itself.

The skills are prompt files. Everything under skills/ is plain text designed to be copied into your own Claude project. When you use them, they run on Anthropic's infrastructure, under your account — not on ours.

• The Python package under src/, and the si-threat-filter and si-topology entry points.
• The FastAPI application under app/, as shipped — for example, a request that crashes it, hangs it, or reads something it should not.
• Packaging and build metadata, including anything that would cause an install to fetch or execute something unexpected.
• The workflows and gate scripts in this repository, particularly anything that would let untrusted input influence what CI executes.
• The skill and ontology files published here, if their content could cause harm to someone who copies them as instructed.

• Anything that requires us to be hosting a service. We are not.
• The retired domains above, and the separate website that replaced them.
• Third-party platforms the project merely names or interoperates with.
• Known vulnerabilities in dependencies that already carry an upstream advisory — please report those upstream. Do tell us if our pinning, our lockfile, or a suppression we have written is the reason you are exposed; that part is ours.
• Findings that depend on an attacker already having local access to the machine running the tool. At that point the tool is not the weakest link.

By default, no analysed text leaves your machine.

The one exception is opt-in and explicit: the topology analyser's Anthropic engine (--engine anthropic) sends the text you are analysing to the Anthropic API, authenticated with your API key, which you supply. It is not the default, it does nothing unless you select it, and the maintainers receive nothing. If you would rather nothing left the machine at all, use the default local engine.

This is a small project without a dedicated security team, so please read the following as honesty rather than indifference: we do not offer a response-time guarantee, and there is no bug bounty. Reports are triaged on a best-effort basis. We would still much rather hear from you than not.

If a report is valid, we will fix it on main, publish an advisory through GitHub, and credit you — unless you would prefer we did not, which is entirely your call.

We will not pursue or support action against anyone who reports a finding in good faith through the channel above, who stays within the scope described here, and who avoids privacy violations and data destruction while investigating.

Since there is no hosted service, almost all legitimate testing of this project happens on your own machine, against your own copy. Please keep it that way.