Skip to content

Security: silentspike/project-sentinel

.github/SECURITY.md

Security Policy

Supported Versions

Version Supported
latest Yes

Reporting a Vulnerability

Do NOT open a public issue for security vulnerabilities.

How to Report

Use GitHub's private vulnerability reporting.

Include:

  • Description of the vulnerability
  • Steps to reproduce
  • Affected component (ECS, Cortex Gateway, Dashboard, Sandbox, etc.)
  • Impact assessment (if known)

Response Timeline

Action Timeframe
Acknowledge receipt 48 hours
Initial assessment 5 business days
Fix development Depends on severity
Public disclosure After fix is released

Severity Classification

Severity Examples
Critical Sandbox escape, arbitrary code execution, credential leak
High Agent data exfiltration, self-recognition pattern detection bypass, auth bypass
Medium Information disclosure, DoS, privilege escalation within sandbox
Low Configuration issues, minor information leaks

Scope

In Scope

  • Rust crates (sentinel-ecs, sentinel-bio, sentinel-sandbox, etc.)
  • Cortex Gateway (Go HTTP proxy)
  • Dashboard (Bun + Hono backend, Vanilla JS frontend)
  • Agent sandbox isolation (bwrap, Landlock, cgroups v2)
  • FlatBuffer schema validation
  • Wasm tool runtime (Wasmtime + Extism)

Out of Scope

  • Third-party LLM provider APIs (Claude, Ollama)
  • Upstream dependencies (report to their maintainers)
  • Social engineering attacks
  • Physical attacks on infrastructure

Security Measures

Automated

  • Weekly cargo audit + govulncheck via GitHub Actions
  • CodeQL scanning for Go and TypeScript (weekly + on push)
  • Dependabot for dependency updates (Cargo, Go, npm, Actions)
  • FlatBuffer schema validation on all inputs

Architecture

  • Agent sandbox isolation: bwrap (namespace) + Landlock (FS ACL) + cgroups v2 (resource limits)
  • No eval(), no SQL concatenation, no innerHTML
  • API keys never logged (redaction in Cortex Gateway)
  • Wasm tools run in isolated Wasmtime sandbox
  • Zero-trust between agents (each agent sees only its own data)

Development

  • Pre-commit hooks enforce cargo fmt + cargo clippy
  • PR review required before merge
  • Conventional commits for audit trail

There aren't any published security advisories