feat(llm-security): playground v7.6.0 fase 5a-d — Tier 3 spesialkomponenter (del 1) [skip-docs]
Integrer fire llm-security-spesifikke Tier 3-komponenter:
- tfa-flow + tfa-leg + tfa-arrow: visualiserer lethal-trifecta-kjede
i toxic-flow-rapport (untrusted-input → sensitive-access → exfil-sink)
- mat-ladder + mat-step: posture-modenhet over kategorier i posture-rapport
- suppressed-group: narrative-audit (v7.1.1) i scan-rapport executive summary
- codepoint-reveal + cp-tag: side-ved-side reveal for Unicode-steganografi
i mcp-inspect-rapport (visible vs decoded)
Endringer:
- Fire nye render-helpers (renderToxicFlow, renderMatLadder,
renderSuppressedGroup, renderCodepointReveal) i hovedscriptet, plassert
før renderScan/Deep/Posture/MCP-Inspect.
- parseScan + parseDeepScan utvidet med narrative_audit-felt via ny
parseNarrativeAudit-helper som ekstraherer "**Suppressed signals:**"-
blokken fra raw_markdown.
- renderScan: meterHtml + suppressedHtml + toxicHtml + owaspHtml + ...
- renderDeepScan: suppressedHtml + toxicHtml + smHtml + matrixHtml + ...
- renderPosture: overall + ladderHtml + smHtml + quickHtml + ...
- renderMcpInspect: invHtml + cpHtml (rebuilt via renderCodepointReveal)
Verifisert:
- tfa-flow=3, mat-ladder=2, suppressed-group=8, codepoint-reveal=12 i HTML
- verdict-pill-lg=20, fp-step=12, scope-security=5 (Sesjon 2-kriterier intakte)
- form-progress__step strict singular=0 (DS canonical bevart)
- Window-globaler intakt (24 unike __-prefiksede globaler)
- JS parse OK (node --check), JSON-state parse OK (3 prosjekter, 18 rapporter)
- HTML-balanse OK (3 script-tags, 1 style-blokk)
- Smoke-test mot demo-data: alle 4 helpers rendrer non-empty HTML med
forventede DS-klasser
Master-plan: plugins/llm-security/playground/V7.6.0-PLAN.local.md (Sesjon 3 av 5).
Sesjon 4 (fase 5e-h: top-risks, recommendation-card, risk-meter, card--severity-*)
neste, deretter Sesjon 5 (verifisering, docs, release).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Kjell Tore Guttormsen
fbda041522