Security

Reporting a vulnerability

Response commitments

These are commitments by Drakon Systems Ltd, not contractual SLAs. We will tell you immediately if a target will slip.

In scope

• The published shieldcortex npm package (latest two major versions)
• The bundled local dashboard server (port 3030)
• The SaaS API at api.shieldcortex.ai (Hono + Drizzle + Postgres on Fly.io)
• The marketing site at shieldcortex.ai (Astro)
• The dedicated OpenClaw plugin @drakon-systems/shieldcortex-realtime
• Defence pipeline correctness issues (e.g. patterns that should fire but don't, false negatives on documented attack classes)
• Credential leak, prompt-injection, and memory-poisoning bypasses against the documented detection layers

Out of scope

• Findings that require a malicious local user already on the host (the package is designed to run on the user's own machine)
• Denial of service against your own local memories.db
• Missing security headers without a demonstrable impact
• Self-XSS in the local dashboard requiring console paste
• Vulnerabilities in better-sqlite3, @anthropic-ai/sdk, Node.js, Fly.io infrastructure, or other third-party software outside Drakon Systems' control — please report those upstream
• Spam, social engineering of staff, physical attacks
• Issues against pre-release branches or unsupported old versions (anything older than the previous major)

Safe-harbour commitment

If you act in good faith — staying within the scope above, avoiding privacy violations and service disruption, and giving us reasonable time to remediate before public disclosure — we will not initiate legal action against you. We extend this protection to:

• Testing against your own machine, your own ShieldCortex Cloud account, or a designated test account you have created
• Authorised testing against teams you own or have written permission to test

We do not authorise testing against the accounts of other ShieldCortex Cloud users.

Engineering posture

• Open source: the npm package and bundled dashboard are MIT-licensed; full source at github.com/Drakon-Systems-Ltd/ShieldCortex
• Reproducible benchmarks: npm run bench regenerates the retrieval scorecard locally; see benchmark/longmemeval/SCORECARD.md
• Defence canary: every shieldcortex doctor run pushes a synthetic injection probe through the firewall and asserts it was blocked (since v4.22.0)
• CI test suite: 1,240+ tests run on every PR; results visible on the public repo
• Transport: TLS 1.2+ enforced on api.shieldcortex.ai and shieldcortex.ai via Fly.io
• Secrets: bcrypt-hashed API keys, single-use 15-minute magic-link tokens, 7-day session tokens, secrets stored in Fly Machines secrets

Compliance roadmap

We publish this honestly rather than claim certifications we do not yet hold.

Enterprise procurement teams that need a specific control: email support@drakonsystems.com and we will respond honestly about feasibility and timeline.

Acknowledgments

Researchers who reported vulnerabilities to us in good faith, with their permission to be named. We do not offer monetary rewards; we do offer credit, a straight answer on triage, and the root cause we found underneath the report.

Operating entity

Drakon Systems Ltd
Company number 16867343 (England & Wales)
Registered office: 34 Lumina Way, Enfield, England, EN1 1FS
support@drakonsystems.com