| Version | Supported |
|---|---|
latest on main |
Yes |
| older releases | No |
Please do NOT open a public GitHub issue for security vulnerabilities.
Instead, use one of these methods:
-
GitHub Private Vulnerability Reporting (preferred): Go to Security Advisories and create a new private report.
-
Email: Send details to alex@kuleshov.tech
- Description of the vulnerability
- Steps to reproduce
- Affected component (e.g.,
ShellTool,FileSystemTool,InjectionGuard) - Potential impact
- Suggested fix (if any)
- Acknowledgement: within 48 hours
- Initial assessment: within 5 business days
- Fix release: depends on severity (critical: ASAP, high: 1-2 weeks)
This project uses multiple layers of protection:
- CodeQL analysis on every push and PR (see
.github/workflows/codeql.yml) - Dependabot alerts and automated security updates
- Secret scanning with push protection enabled
- SpotBugs + PMD static analysis in CI (
mvn verify -P strict) - Input sanitization via
InjectionGuardandInputSanitizer - Sandboxed tool execution with path traversal and command injection protection
- Rate limiting and user allowlists for Telegram access
- Tool confirmation workflow for dangerous operations
The following are in scope for security reports:
- Prompt injection bypassing
InjectionGuard - Path traversal in
FileSystemToolorStoragePort - Command injection in
ShellTool - Authentication/authorization bypasses
- Secrets exposure
- Dependency vulnerabilities (CVEs)
Out of scope:
- Denial of service via LLM API rate limits (upstream provider issue)
- Social engineering attacks on the LLM
- Vulnerabilities in third-party services (Telegram, OpenAI, Anthropic)