Skip to content

Security: cybertechsoft/loopgrid

Security

SECURITY.md

LoopGrid v0.8 Security Posture

LoopGrid v0.8 is a controlled design-partner release, not Production GA. This document describes implemented controls and explicit trust boundaries; it is not a certification or legal compliance statement.

Implemented controls

  • Workspace-scoped service API keys stored only by SHA-256 digest.
  • Service-key scopes: ingest, read, review, admin.
  • Key expiry metadata, revocation and replacement-key rotation.
  • Human session authentication with scrypt password hashing.
  • Workspace roles: owner/admin/reviewer/viewer.
  • Production-mode startup guard requiring PostgreSQL, auth-required mode and strong bootstrap/platform-admin secrets.
  • Configurable CORS with wildcard rejected in strict production mode.
  • Request-size limit and in-process write rate limiting.
  • Request IDs and security response headers.
  • SHA-256 content commitments and workspace-scoped hash chain.
  • Persistent Ed25519 signing in the default design-partner topology.
  • Signed checkpoints.
  • Evidence Bundle v2 with offline verification and optional signer pinning.
  • AES-256-GCM encrypted FULL-disclosure vault separated from the signed ledger.
  • Retention/erasure path that removes encrypted disclosures without rewriting signed commitments.
  • Lifecycle enforcement for consequential event ordering.
  • MCP proxy restricted to configured upstream aliases; arbitrary target URLs are rejected.

PostgreSQL chain append behavior

On PostgreSQL, the ledger serializes workspace chain-head appends through database locking before selecting the predecessor. This is intended to prevent normal same-workspace writers from legitimately creating parallel children of one chain head. The v0.8 design-partner release does not claim a fully validated multi-instance/horizontal-scaling topology until a dedicated multi-instance gate is run.

Platform bootstrap secret vs service identity

LOOPGRID_PLATFORM_ADMIN_KEY is a deployment bootstrap/administrative secret. It is not accepted as a normal evidence read/ingest/review credential. Normal integrations should use workspace-scoped service keys or human bearer sessions.

LOOPGRID_BOOTSTRAP_TOKEN is used only to create the first owner account in production mode; the endpoint becomes unavailable after the first user exists.

Signing trust boundary

Default v0.8 design-partner deployments use a persistent local Ed25519 key. This provides cryptographic signatures and tamper evidence, but it is not represented as a hardware-backed production key-management boundary.

An AWS KMS asymmetric signing path exists separately. Do not claim live hardware-backed signing until the actual IAM/KMS boundary has been exercised in the target deployment.

Timestamping

RFC3161 support validates timestamp imprint/nonce capture and can preserve timestamp.tsr in an evidence bundle. Trusting the TSA signer certificate chain is a separate assertion and should only be claimed after explicit CA-chain validation.

Secrets

For Docker design-partner deployments use:

python scripts\generate_pilot_env.py

The generated .env contains secrets and is gitignored. Do not paste or commit it. Rotate credentials before transferring a deployment to another organization.

Logging

Do not add raw API keys, bearer sessions, provider API secrets or decrypted customer disclosures to application logs. Request IDs, key IDs/prefixes and evidence IDs are preferred for troubleshooting.

Known/deferred hardening

  • live managed/hardware-backed signer boundary;
  • TSA signer certificate-chain trust for a chosen provider;
  • dedicated multi-instance concurrency validation;
  • complete multi-generation evidence-signing key rotation/anchoring policy;
  • enterprise SSO/SAML;
  • external distributed rate limiter;
  • production monitoring/SIEM integration and SLA;
  • independent penetration test/security audit.

Language / claims

Use signed / tamper-evident evidence, not “immutable.” Evidence integrity can support governance, audit and regulatory workflows but does not itself prove HIPAA, GDPR, EU AI Act, SOC 2 or other compliance/certification.

There aren't any published security advisories