Skip to content

security-controls.md

Latest commit

 

History

History
47 lines (41 loc) · 2.51 KB

security-controls.md

File metadata and controls

47 lines (41 loc) · 2.51 KB
Raw

Security Controls

Status: v0.1.0 foundation controls.

Enforced Now

  • Exact Rust 1.97.1 and current reviewed CI tool pins.
  • Edition 2024, resolver 3, locked dependency graph.
  • No third-party crates, Git sources, or registries.
  • publish = false resolved for every Cargo package.
  • no_std and forbid(unsafe_code) in every active crate.
  • Strict compiler/Clippy lints and a 500-line code-file gate.
  • Exact active-crate dependency graph with adversarial reversal, missing-crate, attribute-removal, and oversized-source tests.
  • Deterministic test request IDs refuse reuse after numeric exhaustion; schema keys and purposes reject ambiguous, control, and oversized input.
  • Unit, doc, policy, release-plan, metadata, and cross-target checks.
  • Mandatory unit, assembled-boundary, real-system/acceptance, and adversarial evidence appropriate to every capability claim.
  • Read-only CI permissions and full-SHA GitHub Action pins.
  • cargo-deny, cargo-audit, SPDX SBOM, and CodeQL default setup policy.
  • Exact-commit owner-operated pentest, recorded scope, date, result, resolved findings, and direct-pass or clean-retest outcome before every tag.
  • Closed declared-release set; distinct annotated predecessor tags must form a strict ancestor chain, and only release-status files may change before the first post-pentest GitHub/CodeQL run.
  • Owner-operated pentest approval follows an owner-confirmed finding/retest loop; the record preserves the green pentested commit and any later CodeQL finding, exact changed paths, changed regression tests, and full-gate PASS.
  • Every newly pushed tag receives a record/target/ancestry validation workflow.
  • EUPL-1.2 internal workspace with no registry publishing path.

Designed Now, Implemented Later

  • Tenant/site isolation and capability-based authorization.
  • Privacy-separated PII and purpose-bound fields.
  • Opaque invitation tokens, replay protection, and idempotent mutations.
  • Application-layer PII encryption through reviewed key-provider boundaries.
  • Append-only tamper-evident audit with independent roots.
  • Automatic credential expiry, revocation, and reconciliation.
  • Outbound-authenticated site cells and encrypted offline journals.
  • Deny-by-default Wasm plugins with bounded resources.
  • SSRF-resistant outbound HTTP and signed webhook replay defense.
  • AI classification, disclosure, typed tools, and no authoritative decisions.

Planned controls are not current product claims. Each becomes active only when its release deliverables, tests, documentation, and pentest evidence exist.