Skip to content

STANDARDS.md

Latest commit

 

History

History
62 lines (51 loc) · 6.34 KB

STANDARDS.md

File metadata and controls

62 lines (51 loc) · 6.34 KB
Raw

kit ↔ security standards — coverage & honest gaps

kit's stance. This is an evidence map, not a compliance attestation — the same principle as kit coverage. It says which secure-software standards kit's shipped, deterministic controls align to, and — just as importantly — which ones kit does not yet take into account. Claiming coverage you don't have is exactly the false-green kit exists to prevent, so the gaps below are stated as plainly as the wins.

1. Standards kit takes into account today

Standard How kit engages it Where
OWASP Top 10 (2025) Per-category mapping of kit's shipped controls docs/OWASP_2025.md
OWASP ASVS 4.0.3 (L2) Deterministic evidence map (auto-verified / gap / manual / n-a) kit coverage [--json]
SLSA Build provenance / integrity framing attestation + docs/AUDIT_ATTESTATION.md
Sigstore / cosign Signature verification of artifacts supply-chain paths
OpenSSF Scorecard Repo-posture scoring in CI .github/workflows/scorecard.yml
CycloneDX + SPDX SBOM emit (incl. --agent toolchain) kit sbom
CVE / CWE / OSV / SARIF Vuln IDs, weakness taxonomy, findings interchange kit scan, findings rule field
OWASP Top 10 for LLM Applications (2025) Evidence map of LLM01–LLM10 → kit controls kit coverage --standard=llm-top10
NIST SSDF (SP 800-218 / 218A GenAI profile) Evidence map of the practices kit speaks to (PO/PS/PW/RV) kit coverage --standard=ssdf
EU CRA SBOM-mandate driver kit sbom rationale
GDPR PII redaction / memory handling redaction + memory layer

2. Standards kit does NOT yet take into account (ranked by fit)

Each of these is a genuine gap. Most are standards kit already has controls for but has not mapped — the cheap, honest fix is to extend the kit coverage evidence-map machine, not to invent new controls.

# Standard Why it fits kit Planned home
1 ✅ NIST SSDF — SP 800-218 / 218A Now mapped at the practice level 218A profiles — kit coverage --standard=ssdf (see §1). done
2 ✅ OWASP Top 10 for LLM Applications (2025) Now mapped — kit coverage --standard=llm-top10 (see §1). done
3 NIST AI RMF (AI 100-1) + Generative AI Profile (600-1) AI-specific risk framework; natural parallel to the OWASP mapping. kit coverage mapping
4 MITRE ATLAS Adversarial-AI TTP matrix (ATT&CK for ML) — the vocabulary for the agent threats kit gates. threat-model cross-ref
5 OpenSSF S2C2F (Secure Supply Chain Consumption Framework) Near-perfect fit for install-triage / slopsquat / consumer-side verification — currently only referenced in passing, not mapped. kit coverage mapping
6 ISO/IEC 42001 (AI management system) Certification track for AI governance (the ISO references in-repo are ISO-8601 dates, not 42001/27001). evidence map (GRC)

Secondary / weaker fit: OWASP SAMM & BSIMM (maturity models, not control checklists); EU AI Act (regulatory, vs the CRA already covered); CIS Benchmarks (hadolint covers the Docker slice); SOC 2 / PCI-DSS / HIPAA (compliance regimes — served indirectly by the kit coverage evidence map rather than mapped directly).

3. Why "map", not "add"

The top ones are highest-value because kit already ships the controls they ask for — write-gate, injection quarantine, install-triage, slopsquat scoring, deterministic gates. What was missing was the crosswalk from control → standard clause, exactly what kit coverage already does for ASVS. Status:

  1. ✅ Done: kit coverage --standard=llm-top10 and --standard=ssdf — pinned, curated maps bucketed the same way (auto / gap / manual / n-a), built on a generic evidence-map engine (src/coverage/standard.ts). Control IDs were verified (a source-run confirmed the LLM01–LLM10 titles and the PO/PS/PW/RV practices) before mapping — no invented clause numbers. Caveat carried into the disclaimer: IDs were confirmed via search index, not first-party fetch (egress-blocked), and kit does not enumerate 218A's specific AI task augmentations.
  2. Next: add S2C2F and NIST AI RMF as further evidence maps (same engine).
  3. Every mapping stays behind the "no false green" rule — a clause kit can't confirm is not cited.

This doc is the honest record: kit aligns to the §1 set (now including OWASP LLM Top 10 + NIST SSDF) and does not yet account for the remaining §2 set.