SmartPerfetto processes Android performance traces, which may contain sensitive device information (app package names, system calls, kernel events). We take security and privacy seriously.

If you discover a vulnerability, please do not file a public GitHub issue. Instead:

Preferred: Use GitHub's private security advisory reporting.

Alternative: Email smartperfetto@gracker.dev with:

• A description of the vulnerability and its impact
• Steps to reproduce (a minimal proof of concept helps)
• Your name / handle for attribution (optional)

The in-product Agent-assisted feedback flow follows the same boundary. Marking a candidate as security-sensitive prevents a public draft and routes the user to the private advisory path. SmartPerfetto never submits the report automatically.

If a vulnerability is actively being exploited, we may publish a patched release before the full disclosure window elapses.

In scope:

• SmartPerfetto backend (backend/)
• SmartPerfetto Perfetto plugin (perfetto/ui/src/plugins/com.smartperfetto.AIAssistant/)
• Skill system (backend/skills/)
• CI/CD workflows (.github/workflows/)
• Shell scripts in scripts/ and backend/scripts/

Out of scope:

• The upstream Perfetto project (perfetto/ submodule root) — please report upstream Perfetto issues to google/perfetto.
• Third-party dependencies — please report to the respective maintainers.
• Social engineering, physical attacks, DoS against our infrastructure.

SmartPerfetto is pre-1.0 and under active development. Security patches target main; older tags receive fixes only on request.

• No secrets in repo: API keys, tokens, and credentials live in .env files (gitignored) or GitHub Secrets.
• LLM trust boundary: User-provided trace data and AI-generated content never flow into SQL statements or shell commands without validation.
• Sandboxed trace processing: trace_processor_shell runs as a subprocess with restricted HTTP RPC ports (9100–9900).
• Input redaction: Logged request bodies and device properties go through sanitizeLogData before persistence (see backend/src/services/sessionLogger.ts).
• Public-artifact boundary: Agent-assisted GitHub drafts and local contribution bundles use the shared public-artifact sanitizer. Drafts reject structural prompt attacks and redact credentials, emails, device/network identifiers, absolute paths, and package identities before preview.
• Validated Agent output: External-issue triage runs without tools, accepts only a bounded JSON contract with real source-run references, and falls back deterministically when provider pinning or validation fails.
• No GitHub write credential: The feedback assistant builds an HTTPS issues/new URL only. It accepts no GitHub token and calls no GitHub API.

We will credit reporters in release notes unless requested otherwise. Thank you for helping keep SmartPerfetto safe.