Skip to content

Security: ARPAHLS/OPSIE

Security

SECURITY.md

Security policy

Supported versions

Security updates, when available, apply to the latest tagged revision in this repository lineage maintained by ARPA Hellenic Logical Systems.

Reporting a vulnerability

Please do not file public issues for undisclosed vulnerabilities.

Email: security@arpacorp.net

Include:

  • A concise description of the issue and its impact
  • Steps to reproduce (proof-of-concept if possible)
  • Affected files or subsystems
  • Whether you believe the issue is already exploitable in default configuration

We aim to acknowledge receipt within several business days. Coordinated disclosure is preferred.

Known high-risk areas in this codebase

Review these carefully before any internet-facing deployment:

  1. Environment files: .env must remain untracked. If credentials were ever committed, rotate them everywhere they were reused.
  2. kun.py: Contains database parameters, mail, wallet metadata, and soul signatures. Treat as confidential for production profiles.
  3. AGENT_PRIVATE_KEY: Grants on-chain asset control. Compromise is irreversible without key rotation and fund migration.
  4. web3_handler.py: Missing required Web3 variables causes exit() at import time (availability characteristic, not an access-control boundary).
  5. Biometric authentication: Still-image face matching is not liveness-proof. DeepFace emotion classes are coarse heuristics, not clinical instrumentation. For a standalone gateway focused on facial recognition and emotion policy (and room to add liveness and service boundaries), evaluate Gatekeeper alongside this monolith’s in-process boot gate.
  6. /read: Resolves operator-supplied paths on the host. Only open paths you trust.
  7. Hugging Face /imagine: Requires HUGGINGFACE_API_KEY or HF_TOKEN in .env. Do not embed inference tokens in source.

Safe defaults for operators

  • Run on dedicated hardware with disk encryption.
  • Use least-privilege PostgreSQL roles per user profile.
  • Maintain offline backups separate from live keys.
  • Audit outbound network destinations (Ollama, ElevenLabs, OpenAI, Google, Yahoo Finance, NCBI, Hugging Face, RPC providers).

Credential rotation (incident checklist)

If a repository copy was public with real .env or kun.py secrets:

  • Rotate all API keys, SMTP passwords, Web3 keys, and database passwords that appeared in the leak.
  • Invalidate Hugging Face, OpenAI, Google, ElevenLabs, and explorer API tokens.
  • Regenerate wallet material if a private key was exposed.

There aren't any published security advisories