Skip to content

Security: hanley-development/Agent-Portability-Kit

Security

docs/SECURITY.md

Security Model

Trust boundaries

Repositories, archives, configuration files, skill sidecars, and generated output are untrusted data. The core parser does not execute source content, resolve environment variables, start MCP servers, install dependencies, or contact the network.

Required controls

  • Dry-run is the default.
  • Writes are repository-local and validated before execution.
  • User-authored files require exact-path approval.
  • Raw secrets are excluded from portable models, diagnostics, output, audits, and packages.
  • Archive members are inspected before extraction. Traversal, absolute paths, links, duplicate destinations, special files, excessive sizes, and suspicious compression ratios are rejected.
  • Wrappers validate variable presence without printing values or loading implicit user files.
  • Global configuration, registry, profiles, and home directories are never mutated by core commands.

Report suspected secret exposure or unsafe archive behavior privately to the maintainers before publishing details.

There aren't any published security advisories