MOSA reads local creative assets, Prompt and provenance records, and selected local tool-session data. Security reports are handled privately.
| Version | Security updates |
|---|---|
main and the latest release |
Supported |
| Older tagged snapshots | No planned updates |
v0.1.0 is the final MIT-licensed source snapshot. Later releases use the license in the root LICENSE.
Do not open a public Issue for a suspected vulnerability.
Use one of these private channels:
- GitHub private vulnerability reporting
- Email fengseekling@gmail.com with the subject
[MOSA Security]
Include the affected version or commit, impact, reproduction steps, and a minimal proof of concept. Remove real Prompts, user assets, Tokens, cookies, credentials, local absolute paths, and other private data before sending.
The maintainer will acknowledge the report, assess its scope, and coordinate disclosure and a fix when applicable. Do not publish technical details before a fix or agreed disclosure date.
- MOSA is designed for loopback use on
127.0.0.1. Public ports, reverse proxies, shared-host deployment, and untrusted multi-user access are unsupported. - Web Capture is disabled unless
MOSA_WEB_CAPTURE_TOKENis explicitly configured. Treat the Token as a local secret. - Browser-extension origins, input envelopes, declared MIME types, decoded image bytes, file sizes, and pixel counts must be validated before ingest.
- Codex, Grok, and Cowart source roots and Cowart insertion targets are explicit allowlists. Do not widen them to work around an import failure.
- Imported media, Prompt text, provenance, page URLs, and session identifiers are private user data by default.
See PRIVACY.md and docs/operations.md for the data and deployment boundaries.
- Vulnerabilities in Codex, Chrome, ChatGPT, Grok, Cowart, tldraw, Node.js, or operating-system components should be reported to their respective maintainers.
- Findings that require exposing MOSA to a public interface, disabling documented validation, or granting an attacker local filesystem access are outside the supported deployment model, though defense-in-depth reports are still welcome.
- This project does not currently operate a paid bug-bounty program.