You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Introduces a cross-cutting modifier (L0-L4) that describes how
heavily the software integrates LLMs at runtime, separate from
the 5 code dimensions. L3+ pushes the effective tier floor:
- L3 (Tool Use) → min Tier 3
- L4 (Agentic) → min Tier 4
At L3+, a callout points to specialized frameworks (OWASP LLM
Top 10, Palo Alto SHIELD, Aikido VCAL, Google SAIF) since the
built-in mitigation catalog covers build-time risks only.
Phase 1 (data model + tier logic):
- getTierIndex(values, llmRuntimeLevel) with hard floor multiplier
- llmRuntimeLevel state in RiskRadar, passed to RadarChart
- RadarChart's rAF updater reads level via ref (closure-safe)
Phase 2 (minimal UI):
- Pill-button row between chart and tier badge
- Callout box at L3+ with framework links
- DE/EN i18n for labels, level descriptions, callout text
Phases 3-5 (presets, extended docs, skills integration) follow.
Refs #20
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
0 commit comments