Skip to main content

Security

Dobby Security

jhf-dobby is the Fabric-governed adaptive-learning module for Helpifyr. It produces recommendation and learning posture, but it must not become an unreviewed authority for external actions, tenant-crossing synthesis, policy changes, or source-code mutation.

Security Boundaries​

  • Dobby must fail closed to warmup, observe_only, or proposal_only when Fabric truth, approval integrity, policy integrity, or tenant separation is missing.
  • No-learn zones are hard boundaries and must not be bypassed by replay, promotion, or recommendation flows.
  • Learning output is advisory unless an admitted owner workflow explicitly promotes it.
  • Dobby must not mutate external repositories, workflow files, source files, or production systems directly.
  • Sensitive evidence should be linked by owner reference, not copied into public docs, Gitea comments, logs, or generated recommendation text.
  • Runtime diagnostics must be bounded and must not leak secrets, tokens, credentials, tenant data, or private trace payloads.

Public Documentation Safety​

Public manufacturer docs may describe Dobby's advisory role, tenant separation, no-learn posture, and verification paths. They must not expose private learning records, raw traces, tenant-specific samples, or operator-only evidence.

Verify Path​

  • python scripts/validate_docs_inventory.py
  • python scripts/validate_docs_platform_v1_6.py
  • python scripts/validate_module_docs.py

License​

AGPLv3. See LICENSE. Project: https://helpifyr.com