This is a personal open-source project. If you discover a security vulnerability in Kaya / ai-assistant, please follow responsible disclosure:

Preferred method: GitHub Security Advisories

Use GitHub's private vulnerability reporting to disclose security issues confidentially:

1. Go to https://github.com/jstilb/ai-assistant/security/advisories/new
2. Fill in the advisory form with details of the vulnerability
3. Submit — this creates a private advisory visible only to the maintainer

Alternative method: Email

If you prefer email contact, send details to the address listed in the GitHub profile: https://github.com/jstilb

A useful vulnerability report includes:

• Description: What the vulnerability is and what it could allow an attacker to do
• Steps to reproduce: Minimal steps to demonstrate the issue
• Impact assessment: What data or systems are at risk, and under what conditions
• Suggested fix (optional): If you have a proposed mitigation

This is a personal project maintained by one person. Response times:

1. The maintainer acknowledges receipt and assesses severity
2. A private advisory is created (or already exists from your submission)
3. A fix is developed and tested
4. The fix is merged to main
5. A public advisory is published after the fix is deployed
6. Credit is given to the reporter in the advisory (unless anonymity is requested)

In scope for vulnerability reports:

• Prompt injection vulnerabilities in the hook pipeline (PromptInjectionDefender, SecurityValidator)
• Secret exposure via logging, error messages, or unintended file access
• Permission model bypass — operations that should be blocked but aren't
• Security configuration issues that weaken the defense posture
• Vulnerabilities in skill code that could be exploited via the hook system

Out of scope:

• Vulnerabilities in Claude Code itself (report to Anthropic)
• Vulnerabilities in Bun runtime (report to Bun team)
• Social engineering attacks against the user
• Issues that require physical access to the machine
• The fact that this is a personal AI system with broad system access (this is by design)

For a complete description of the security architecture, threat model, and defense layers, see:

• docs/SECURITY.md — Full security architecture documentation
• docs/HOOK-PIPELINE.md — Hook pipeline internals including PromptInjectionDefender design

Responsible disclosure is appreciated. Contributors who discover and responsibly report security vulnerabilities will be acknowledged in the advisory and in the project's release notes.