Skip to content
This repository was archived by the owner on Jun 19, 2026. It is now read-only.

Security: Lilith-Systems/invite

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
0.1.x
< 0.1.0

Only the latest minor release of the 0.1.x series receives security updates.

Reporting a Vulnerability

Please do not file public issues for security vulnerabilities.

Report security issues privately to the project owner at: security@localhost (replace with actual contact)

Include the following information:

  • Description of the vulnerability
  • Steps to reproduce or proof-of-concept
  • Affected versions
  • Potential impact assessment
  • Any suggested mitigations

We aim to acknowledge reports within 48 hours and provide a remediation timeline within 5 business days.

Security Model

The NVIDIA Gratitude Driver is a user-mode telemetry and routing advisor with the following safety guarantees:

Process Isolation

  • ✅ No kernel hooks or kernel-mode code
  • ✅ No DLL injection into other processes
  • ✅ No game memory reads or writes
  • ✅ No screen capture or pixel reading
  • ✅ No ptrace, debug registers, or debugger attachment

Data Access

  • ✅ Reads only aggregate GPU telemetry via NVML / nvidia-smi
  • ✅ Reads only aggregate Chrome process counts, working set totals, and command-line types
  • ❌ Does not read URLs, page content, cookies, credentials, browsing history, or user input
  • ❌ Does not read browser local storage, IndexedDB, or extension data

Network & API

  • ✅ No outbound network connections initiated by the driver
  • ✅ Does not bypass API limits or provider controls
  • ✅ Does not rotate accounts, automate quota abuse, or circumvent rate limits
  • ✅ Provides backoff recommendations to respect provider limits

Runtime Telemetry

Runtime telemetry (runtime/nvidia_gratitude_driver/status.json and telemetry.jsonl) may reveal:

  • Local hardware utilization (GPU VRAM, utilization, temperature, power)
  • Aggregate Chrome process counts and memory working sets
  • Route decisions (LOCAL_CEREBELLUM / HYBRID / CLOUD_CORTEX)

Store telemetry logs according to your organization's data retention and privacy policies.

Threat Model (STRIDE Summary)

Threat Category Mitigations
Spoofing No authentication surface; driver does not accept external commands
Tampering Atomic JSON writes with .tmp replace; append-only JSONL log rotation; no config reload from untrusted paths
Repudiation Telemetry is local-only; no audit log forwarding without operator action
Information Disclosure Only aggregate GPU and process telemetry exposed; no PII or browser content
Denial of Service Bounded log rotation; EWMA smoothing prevents flapping; cooldown prevents thrash
Elevation of Privilege Runs as standard user; no setuid, no capabilities, no admin required

Supply Chain

  • Dependencies: nvidia-ml-py and psutil (pinned in pyproject.toml)
  • CI runs pip-audit on every push and PR
  • SBOM generated at release via cyclonedx-python
  • Release artifacts signed with cosign (sigstore); provenance attestations published to GitHub Releases

Hardening Checklist

  • No elevated privileges required
  • No capabilities retained
  • No sudo/admin needed for any operation
  • Input validation on terminal event parsing (regex bounded)
  • Subprocess calls use explicit argv, no shell=True
  • File writes use atomic replace pattern
  • Log rotation bounds disk usage
  • Dependencies scanned for CVEs in CI

Contact

For security questions or responsible disclosure coordination, contact the project maintainer.

There aren't any published security advisories