Skip to content

Security: veritaswiki/skill-evolution

Security

SECURITY.md

Security Policy

Safety Invariants

skill-evolution enforces strict security boundaries by design:

  • Read-Only Review Fork: The review fork cannot write files or execute shell commands directly; it emits structured JSON validated by lib/gatekeeper.js.
  • Provenance Isolation: Skills without metadata.createdBy: "agent" are treated as user-owned and are never auto-modified.
  • No Destructive Deletes: Archiving moves skills into .archive/; every mutation is recorded in .evolution_ledger.jsonl with content-addressed SHA-256 blobs and fail-closed rollback.
  • No bundled endpoint or credentials: the package ships no gateway address and no keys. The optional auxiliary review channel is only used when DSH_EVOLUTION_BASE_URL (plus DSH_EVOLUTION_API_KEY / DSH_EVOLUTION_MODEL) is set by the user.
  • Local HTTP API is unauthenticated: the workbench API and the fixed-port mobile page (default 18890, DSH_EVOLUTION_PORT) are meant for local or trusted-network use. Do not expose them to the public internet.

Reporting a Vulnerability

Please open a security advisory on GitHub at https://github.com/veritaswiki/skill-evolution/security/advisories.

There aren't any published security advisories