feat(llm-security): playground Fase 3 — v7.5.0 med 18 parsere/renderere

Single-file SPA playground har nå parser + renderer for alle 18
produces_report=true-kommandoer (Fase 2: 10 høy-prio + Fase 3: 8
gjenstående: mcp-inspect, supply-check, pre-deploy, diff, watch,
registry, clean, threat-model). 18 markdown test-fixtures fungerer
som kontrakt-anker for parser-utvikling.

Komplett demo-prosjekt `dft-komplett-demo` har alle 18 rapporter
ferdig parsed inline — klikk-gjennom uten "parser ikke implementert"-
paneler. 2 nye archetypes i KEY_STATS_CONFIG: kanban-buckets (clean)
og matrix-risk (threat-model).

Bug-fix: normalizeVerdictText sjekker nå GO-WITH-CONDITIONS /
CONDITIONAL / BETINGET FØR plain GO så betinget verdict (pre-deploy
med åpne vilkår) ikke kollapser til ALLOW.

Eksponert 11 window-globaler for testing/automasjon (__store,
__navigate, __loadDemoState, __PARSERS, __RENDERERS, __CATALOG,
__inferVerdict, __inferKeyStats, __renderPageShell,
__handlePasteImport, __scheduleRender). 12 Playwright-genererte
screenshots i playground/screenshots/v7.5.0/.

A11Y-rapport (WCAG 2.1 AA): 0 blokkerende, 3 mindre forbedringer
flagget for v7.5.x patch (skip-link, heading-hierarki på project,
aria-live toast).

Versjonsbump 7.4.0 -> 7.5.0 i 10 filer (package.json, plugin.json,
CLAUDE.md header, README badge, CHANGELOG-entry, 3 scanner VERSION-
konstanter, ROADMAP, marketplace-rot README).

Ingen scanner- eller hook-behavior-changes — purely additive surface.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

plugins/llm-security/playground/test-fixtures/audit.md

@@ -0,0 +1,141 @@
+# Full Security Audit — DFT marketplace
+
+---
+
+## Header
+
+| Field | Value |
+|-------|-------|
+| **Report type** | audit |
+| **Target** | ~/repos/dft-marketplace |
+| **Date** | 2026-05-05 |
+| **Version** | llm-security v7.4.0 |
+| **Scope** | 7 audit dimensions, 10 OWASP categories |
+| **Frameworks** | OWASP LLM Top 10, OWASP Agentic |
+| **Triggered by** | /security audit |
+
+---
+
+## Risk Dashboard
+
+| Metric | Value |
+|--------|-------|
+| **Risk Score** | 31/100 |
+| **Risk Band** | Medium |
+| **Grade** | C |
+| **Verdict** | WARNING |
+
+| Severity | Count |
+|----------|------:|
+| Critical | 0 |
+| High | 4 |
+| Medium | 8 |
+| Low | 7 |
+| Info | 9 |
+| **Total** | **28** |
+
+**Verdict rationale:** Posture base grade B downgraded to C after agent-level findings (4 high). No critical, but `Logging & Audit` and `Permission Hygiene` need attention.
+
+---
+
+## Executive Summary
+
+Full audit combined posture-scanner output with skill-scanner-agent and mcp-scanner-agent narratives. 28 findings across 14 files. Most concentrated in agent definitions (over-permissioned tool lists) and `.claude/settings.json` (missing audit log + wildcard Bash). Recommendation: address top 3 actions to reach Grade B; six more to reach Grade A.
+
+---
+
+## Radar Axes
+
+| Axis | Score |
+|------|------:|
+| Deny-First Configuration | 4 |
+| Hook Coverage | 5 |
+| MCP Trust | 3 |
+| Secrets Management | 5 |
+| Permission Hygiene | 2 |
+| Supply-Chain Defense | 4 |
+| Logging & Audit | 1 |
+
+---
+
+## Category Assessment
+
+### Category 1 — Deny-First Configuration
+
+| Status | PASS |
+
+**Evidence:** `.claude/settings.json` has `permissions.defaultMode: "deny"`. Explicit allow-list in place.
+
+**Recommendations:** None — Grade A on this axis.
+
+### Category 2 — Hook Coverage
+
+| Status | PASS |
+
+**Evidence:** 9 hooks active (PreToolUse: 4, PostToolUse: 2, UserPromptSubmit: 1, PreCompact: 1, others: 1).
+
+**Recommendations:** Consider adding PreCompact-poisoning detection if not already covered.
+
+### Category 5 — Permission Hygiene
+
+| Status | PARTIAL |
+
+**Evidence:** 3 agents have `Write` in tool list. 1 has `Bash` without sub-command restriction.
+
+**Recommendations:** Tighten tool lists to minimum-necessary set. Use `Bash(git:*)` instead of `Bash(*)`.
+
+### Category 11 — Logging & Audit
+
+| Status | FAIL |
+
+**Evidence:** No `audit.log_path` configured. No SIEM integration. No JSONL audit-trail.
+
+**Recommendations:** Enable `audit.log_path` immediately — closes 1 high + 3 medium findings.
+
+(Categories 3, 4, 6-10, 12-13 follow same format — see envelope JSON for full breakdown)
+
+---
+
+## Risk Matrix (Likelihood × Impact)
+
+| Category | Likelihood | Impact | Score |
+|----------|-----------:|-------:|------:|
+| Logging gap (PST-001) | 4 | 4 | 16 |
+| Permission sprawl | 3 | 4 | 12 |
+| MCP drift (airbnb-mcp) | 3 | 3 | 9 |
+| AI Act classification missing | 2 | 3 | 6 |
+
+---
+
+## Action Plan
+
+### IMMEDIATE (this week)
+
+1. Enable audit-trail: set `audit.log_path` in `.llm-security/policy.json`
+2. Tighten 3 over-permissioned agents (drop `Write` where unused)
+3. Investigate airbnb-mcp drift — reset baseline only after review
+
+### HIGH (this month)
+
+4. Document AI Act risk classification in `CLAUDE.md`
+5. Replace `Bash(*)` with `Bash(git:*, npm:*)` in `.claude/settings.json`
+6. Bump 2 dependencies to clear OSV advisories
+
+### MEDIUM (next quarter)
+
+7. Add SECURITY.md disclosure policy
+8. Trim verbose skill descriptions (3 files)
+9. Document hook rationale in plugin CLAUDE.md
+
+---
+
+## Positive Findings
+
+- All hooks active and non-bypassed
+- No critical findings
+- Posture scanner runtime < 2s (well-tuned)
+- Memory hygiene clean
+
+---
+
+*Audit complete. 28 findings, Grade C, 14.7 seconds.*