SentinelFlow is an independent, educational, open-source portfolio project. It is not a bank system, not a regulatory compliance product, and not a production fraud-decision engine. It runs on synthetic data only and makes no real financial decisions.

The local stack is designed to bind to a developer's own machine. It is not a deployment target and has never been hardened for one. Do not expose it to a network you do not control, and do not put real data into it.

The project is pre-1.0. There is no long-term support branch, and there is no commitment to backport a fix to an earlier tag.

Do not open a public issue for a security problem.

Report it privately through GitHub:

https://github.com/la3679/sentinelflow/security/advisories/new

Please include:

• What the problem is, and which component it affects
• How to reproduce it, ideally from a clean clone
• What an attacker could achieve
• The commit or tag you tested

This is a personal project maintained by one person in their own time. It has no on-call rotation and no service-level agreement, and it would be dishonest to publish response times it cannot meet. What is committed to:

• An acknowledgement that the report was received and read.
• An honest assessment: accepted, not a vulnerability, or accepted but out of scope for a project of this kind — with the reasoning either way.
• Credit in the advisory and the changelog, unless you prefer otherwise.
• Coordinated disclosure: the advisory is published once a fix is available, or once it is clear no fix will be made.

• Injection of any kind in apps/api or apps/scoring
• Authorization flaws — a role reaching data or an action it should not
• Secrets or credentials committed to the repository or written to logs, metrics labels, event payloads, or error responses
• Unbounded list, export, or replay endpoints that allow resource exhaustion
• Formula injection in CSV export
• Container escapes or privilege escalation from the images this repository builds
• Supply-chain problems in the build: an unpinned action, a missing checksum, a dependency that is not what it claims to be

• Anything requiring the local stack to be exposed to the public internet. It is documented as local-only and is not built to be deployed.
• The absence of a control the project explicitly does not implement. Role handling in the console is a user-experience affordance and is documented as never being a security boundary.
• Findings against a demonstration credential generated by make bootstrap. Those are random per machine, git-ignored, and protect nothing real.
• Vulnerabilities in a third-party dependency with no available fix. Report those upstream; SentinelFlow's scanners already surface fixable ones and fail the build on them.
• Denial of service against a single-broker, single-node local stack. It is not built for availability.

These are the controls that exist today, not aspirations:

• Test against anything other than your own local instance.
• Include real financial data, real personal data, or credentials in a report.
• Open a public issue, a discussion, or a pull request that describes an unfixed vulnerability.