ms-ai-architect/CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Changed

• OKF second-brain (sporet for brukerens egen kontekst-wiki — IKKE de 389 skill-refs): Ratifiserte den delte katalog-spec-en catalog/docs/okf-second-brain/spec.md v0.1 som eneste kilde for konvensjonen; lokal brief (docs/okf-second-brain-brief-2026-06.md) refererer den nå i stedet for å restate. Premiss-korreksjoner registrert (spec §9): mdcode/kcmd er ikke et OKF-verktøy (droppet fra adopsjonsplanen); ingen gjenbrukbar OKF-ingest-kode finnes (reference_agent er GCP/Gemini-bundet → adopter prompt-mønstre, ikke kode); kanonisk anbefalt feltnavn er resource (ikke source). Full OKF-pakke-ambisjon består — minimal kontrakt (spec §3) er gulvet, rike felt rir over som extension keys (spec §5). Kun docs/plan-justering; bygging er separat go.

[1.17.0] - 2026-07-04

Spor 1 — Port-1-substrat migrert på korpuset (de 4 ikke-advisor-skills) via den deterministiske migrate-corpus.mjs-applieren. Reference-kontrakten (**Type:** / **Source:** / TOC) er nå påført på tvers av korpuset, og stale **Verified:** MCP-poison er fjernet. Ingen kommando-, agent- eller persona-endringer; prosa-kropper byte-identisk; advisor-skillen urørt.

Changed

• Reference-headere (327 filer): **Type:** stemplet på alle 327 (reference/methodology/regulatory/template); **Source:** (Microsoft Learn-autoritet, satt per fil i manifestet) backfilt på 243 sourced reference-filer; ## Innhold TOC innsatt på 325 store filer (≥100 linjer). verified forblir null — settes først i en senere corpus-judge-pass.
• Full-pass-worklist aktivert: 243 due, 0 fresh. CT5 (referanse-kontrakt) available på alle 4 skills; N4 (TOC) 0 → 1.000/pass. Skill-scorer steg (95→98–100); den forhåndsdeklarerte «transient sub-90 dip» inntraff ikke (registrert i utfallsdoc).

Fixed

• Stale-verified poison: **Verified:** MCP <dato> fjernet — 14 header-forekomster + 9 stray body-duplikater innenfor 500-byte-scan-vinduet (mlops-genaiops). Disse ble ellers falskt lest som «verified» og droppet fra worklisten. Kun stray metadata-linjer fjernet, aldri prosa.
• insertHeaderFields anker-overshoot: når en meta-linje selv passerer 500 B (2 filer med et avsnitt pakket i **Status:**), faller ankeret nå tilbake til siste meta-linje innenfor scan-vinduet, så Type/Source forblir parsbare. TDD.
• normalizeStaleVerified: fjerner nå alle stale non-date **Verified:** i 500B-vinduet uansett ----posisjon (operatør-godkjent utvidelse av carve-out). TDD.

Suite 728/728 grønn. Detaljert utfall + policy-delta + base-felt-restanse: docs/spor1-migration-outcome.md.

[1.16.5] - 2026-06-24

KB-refresh-patch. LOW-tier (52 reference docs) verifisert mot Microsoft Learn (post-1.16.4 baseline). Ingen kommando-, agent- eller skill-endringer — kun reference docs. ~33 filer var TRYGT (innhold gjeldende, kun header-bump); resten hadde reell drift, URL- eller syntaksfiks. Hver VERIFY-påstand re-bekreftet av main-agent mot sitert Microsoft Learn-kilde FØR skriving; git diff + plugin-validering (239/0). Needing-update 52 → 0 (alle tiers nå 0).

Changed

• Agent Framework (materiell fiks): agent-framework fullstendig omskrevet — tidligere utgave bygde på et oppdiktet API (azure.ai.agent, FoundryClient/FoundryTools, Swarm/Handoff, Memory, @checkpoint, enable_tracing). Korrigert til det ekte SDK-et agent-framework (Python) / Microsoft.Agents.AI (.NET): Agent(client=...), @tool, AgentSession, graf-baserte WorkflowBuilder/SequentialBuilder, OpenTelemetry-observability, og den faktiske Foundry-verktøykatalogen. Etterfølger til BÅDE Semantic Kernel og AutoGen; .as_agent_framework_tool-bro (krever semantic-kernel ≥ 1.38). migration-patterns §6 tilsvarende korrigert (Agent Framework = eget SDK, ikke et SK-namespace).
• Regulatorisk (norsk AI Act): public-sector-checklist + ai-utredning-template — norsk ikrafttredelse mykgjort fra «aug 2026 samtidig med EU» til «forsinket pga. EØS-tilpasning, ventet 2026»; Foundry URL-slug ai-studio → foundry.
• Copilot/lisens: Copilot Studio «billed sessions» → Copilot Credits (felles valuta fra 1. sep 2025) i migration-patterns og licensing-matrix.
• Data engineering (Fabric/Databricks): zero-etl-fabric-patterns Oracle/SAP mirroring Preview→GA; onelake-data-strategy OneLake Security + Shortcut Transformations GA; lakehouse-architecture-design liquid clustering GA; data-quality-ai-frameworks MLV ON MISMATCH-syntaks + DLT→Lakeflow SDP-rebrand.
• API Management (GenAI gateway): genai-gateway-policies / logging-analytics-ai-traffic / security-hardening-ai-gateway — token-quota-period, KQL ModelDeploymentName → DeploymentName, llm-token-limit ikke støttet på Consumption, semantic-cache / Developer Portal v2-tier-matriser korrigert, llm-content-safety policy-syntaks.
• BCDR: state-management-failover Azure Cache for Redis retirement → Azure Managed Redis; multi-region-azure-openai-deployment + rto-rpo-planning-ai-services BCDR-ref-URLer fikset.
• Cosmo-utfasing (start): «For Cosmo»-seksjonene fjernet i agent-framework og disconnected-ai-scenarios (persona-pedagogikken er under avvikling — full fjerning på tvers av repoet er planlagt som eget initiativ).
• Header-bump (TRYGT, ~33 filer): data-engineering-, api-management-, bcdr- og advisor-klyngene (innhold verifisert gjeldende mot Microsoft Learn).

[1.16.4] - 2026-06-24

KB-refresh-patch. MEDIUM-tier (33 reference docs) verifisert mot Microsoft Learn (post-1.16.3 baseline). Ingen kommando-, agent- eller skill-endringer — kun reference docs. 18 filer var TRYGT (innhold gjeldende, kun header-bump); 15 filer hadde reell drift. Hver VERIFY-påstand fra de 7 parallelle drift-rapportene ble re-bekreftet av main-agent mot sitert Microsoft Learn-kilde FØR skriving; disclaimed/illustrative priser urørt; GPT-4.1-pricing-påstand forble uverifiserbar (kun community-kilder) og ble IKKE endret; alle endringer review-gated på git diff + plugin-validering (239/0). Medium-count 33 → 0 (needing-update 85 → 52).

Changed

• Prompt-caching: token-optimization — retention-modell omskrevet: in-memory (5–10 min, ≤1t, default t.o.m. gpt-5.4) vs extended retention 24t (prompt_cache_retention, default for nyere modeller); in-memory støttes av «GPT-4o eller nyere».
• Realtime: real-time-reasoning — la til gpt-realtime-1.5 (2026-02-23); Realtime API flyttet til GA-endepunkt /openai/v1 (ikke lenger date-baserte api-version); preview→GA-statusnyanse.
• Guardrails-rebranding: regulatory-and-compliance + azure-ai-foundry — «content filters» → Guardrails (and controls), default Microsoft.DefaultV2, fire intervensjonspunkter; data-privacy-kildetittel → «Models sold by Azure».
• GPU/PTU: gpu-compute-sizing — ND96asr_v4 = 8×A100 40GB = 320 GB (var 640 GB); load-testing — PTU min deployment Global/Data Zone 15 (increment 5), 50 kun Regional.
• Foundry-deprecations: azure-ai-foundry — Deep Research-tool og Connected agents (classic) deprecated (retire 2027-03-31), Prompt Flow-retirement 2027-04-20, A2A-tool nå i Java SDK.
• Agent-evaluering/-memory: agent-evaluation — nye agent-evaluatorer (Task Completion/Customer Satisfaction/Tool Selection m.fl.) + FoundryEvals + gpt-4.1-mini judge; agent-memory — procedural memory (3. type), TTL/CRUD, Norway East-region bekreftet (VNet ikke støttet).
• Øvrig faktafiks: multi-turn RPM/TPM-forhold varierer per modell; structured-output usupporterte keywords utvidet + gpt-4o-mini-audio-preview; temperature-sampling Copilot Studio prompt builder-temp + fjernet uoffisielt M365-tall; rag-hallucination groundedness mitigating/correctionText; model-versioning MLflow stages legacy i UC; speaker-recognition Limited Access-status; chain-of-thought reasoning_effort per-modell-nyanse.
• Header-bump (TRYGT, 18 filer): domain-specific, few-shot, function-calling, grounding, role-playing, system-message-design, async-processing, batch-api-usage, concurrent-request, performance-benchmarking, response-chunking, throughput-optimization, token-per-second, late-chunking, rag-evaluation, self-reflective-rag, streaming-rag, feedback-loops.

[1.16.3] - 2026-06-24

KB-refresh-patch. HIGH-tier (11 reference docs) + 1 dateless kjernefil (ms-ai-advisor/references/architecture/security.md), flagget av endrings-detektoren mot Microsoft Learn-sitemaps (post-1.16.2 baseline), er verifisert og oppdatert. Ingen kommando-, agent- eller skill-endringer — kun reference docs. Hver VERIFY-påstand fra de parallelle drift-rapportene ble bekreftet mot sitert Microsoft Learn-kilde FØR skriving (main-agent som verifiseringsgate); disclaimed/illustrative priser urørt; alle endringer review-gated på git diff + plugin-validering (239/0). High-count 11 → 0, Unverified 1 → 0 (needing-update 97 → 85).

Changed

• Content Safety: bias-detection image-modeller korrigert til Medium default-terskel (var feilaktig «Low»), verifisert mot default-safety-policies; responsible-ai-policy Foundry Agent Service monitoring merket (Preview); content-safety-implementation tegnkode-typo (申请 → Apply for). Avkreftet ved verifisering: Content Safety Analyze Text 10K-tegn-grensen var korrekt (drift-rapportens «1000» var feillesing av Transparency Note-ens nøyaktighets-anbefaling — gaten fanget den).
• Guardrails/agenter: agent-autonomy-and-control-governance la til Task Adherence (preview) i risk-kategoriene (bekreftet i guardrails-overview).
• Evaluering: response-quality-metrics-rag la til Prompt Flow retirement 2027-04-20 (migrer til Microsoft Agent Framework) + korrigerte gpt-4.1-mini (GPT-4.1-variant, ikke o-series reasoning model).
• Opphavsrett: copyright-ai-training-data-norway oppdatert til «Models sold by Azure»-terminologi (var «Azure Direct Models»), la til 5. data-privacy-garanti (ingen forbedring av Microsoft-/tredjepartsprodukter uten samtykke), klargjorde CCC-scope (Copilot Studio + GitHub Copilot = tjeneste-spesifikke mitigations) og oppdaterte metaprompt-referansen til safety-system-message-templates.
• Sikkerhet (kjernefil): architecture/security.md fikk maskinlesbar Last updated-header (var dateless → unverified), «Hate and Fairness»-navn, og CMK Key Vault-krav korrigert (Azure RBAC «Key Vault Crypto Service Encryption User» som primæralternativ, ikke «legacy access policies»). RBAC-rollenavn bekreftet uendret.
• Header-bump (TRYGT, innhold verifisert gjeldende): owasp-llm-top10-azure-mitigations, defender-threat-protection-ai-services, output-validation-grounding-verification, stakeholder-communication-ai-decisions, model-performance-drift-detection.

[1.16.2] - 2026-06-24

KB-refresh-patch. Critical cost-klyngen (8 reference docs i cost-optimization/), flagget av endrings-detektoren mot Microsoft Learn-sitemaps (post-1.16.1 baseline), er verifisert og oppdatert. Ingen kommando-, agent- eller skill-endringer — kun reference docs. Hver VERIFY-påstand fra de parallelle drift-rapportene ble bekreftet mot sitert Microsoft Learn-kilde FØR skriving (main-agent som verifiseringsgate); disclaimed/illustrative priser urørt; alle endringer review-gated på git diff + plugin-validering (239/0). Critical-count for klyngen 8 → 0.

Changed

• Deployment-katalog: Spillover GA (var preview), ny Priority processing-kategori (4. deployment-type) og Data Zone Batch (SKU DataZoneBatch, EU/US-datasone) lagt inn på tvers av cost-filene; modell-lineup GPT-5 → GPT-5.1/5.5.
• Prompt-caching: extended cache retention 24t (prompt_cache_retention) reframet mot in-memory vs extended; cache-TTL-akse korrigert.
• Fine-tuning: RFT (Reinforcement Fine-Tuning) kostnadsmodell lagt til (tidsbasert + $5000 per-job-tak), utover SFT/DPO.
• Batch: «exponential backoff queuing» reframet til server-side fail-fast + klient-side backoff-retry (kodeeksempelet er klient-side).
• On Your Data: deprecation (pensjon 2026-10-14) flagget der relevant.
• Kalibrerings-korreksjoner: fjernet uverifiserbart parametertall (GPT-4o «175B+»); «GPT-4 utfaset» → «legacy» (GPT-4-familien superseded, ikke fjernet); avkreftet én drift-påstand ved verifisering (Phi-3/Phi-2/Falcon-7B fortsatt i Foundry-katalogen → ingen «retired»-merking lagt inn).
• Filer: ptu-vs-paygo-economics, reserved-capacity-planning, cost-monitoring-cost-attribution, inference-endpoint-cost-optimization, prompt-engineering-cost-reduction, token-counting-optimization, request-batching-aggregation, small-language-models-economics.

[1.16.1] - 2026-06-23

KB-refresh-patch. Tre akkumulerte kunnskapsbase-oppdateringer som speiler Microsofts produkt-navngiving og modellkatalog. Ingen kommando-, agent- eller skill-endringer — kun reference docs. Hvert premiss verifisert mot Microsoft Learn FØR sveip; alle endringer review-gated på git diff + plugin-validering + deterministisk re-score (skills uendret, alle ≥90 %).

Changed

• Modellkatalog GPT-5.1 → GPT-5.5 i 3 filer: 2 kostnadsfiler (batch-processing-cost-reduction.md, model-selection-price-performance.md) + plattformfila azure-ai-foundry.md.
• Foundry-navnesveip «Azure AI Foundry» → «Microsoft Foundry» på tvers av 233 reference docs (810 forekomster) — følger Microsofts rebranding av Foundry-produktnavnet.
• RBAC-rolle-renavn «Azure AI User» → «Foundry User» (foundry-workflows-visual-orchestration.md); den eldre rollenavn-mappingen i azure-ai-foundry-cost-governance.md ble bevart med kontekst.

[1.16.0] - 2026-06-22

Currency-, privat-sektor-paritets- og selvstendiggjørings-release. Adresserer devil's-advocate-audit (2026-06-18, 10 dimensjoner, 89 verifiserte funn) — docs/devils-advocate-audit-2026-06-18.md. Alle faktapåstander verifisert mot Microsoft Learn / EDPB / OWASP-primærkilder før skriving.

Added

• 4 nye kommandoer (25 → 29):
  • /architect:businesscase — forretningscase med NNV (netto nåverdi) + DFØ-gevinstrealisering (#7c).
  • /architect:anskaffelse — AI-anskaffelse: kravspesifikasjon, leverandørevaluering, terskelverdier (#7c).
  • /architect:design — sektor-nøytralt Solution Architecture Document (SAD): kontekst, NFR, alternativer, valgt design, risiko, veikart (#7d).
  • /architect:vendor — tredjeparts-/SaaS-leverandørvurdering (due diligence): dataresidens, sub-prosessorer, DPA, Schrems II, AI Act-deployer (#7d).
• 2 nye KB-filer (ms-ai-security, 60 → 62 refs): owasp-llm-top10-azure-mitigations.md (LLM04/06/08/09) og defender-threat-protection-ai-services.md (AI threat protection — ikke i Azure Government).
• Privat-sektor-paritet: sektor-parametrisering i 6 kommandoer + onboarding-forgrening for privat/regulert sektor + parallell privat-bane i README/help/playground.
• Spor C — selvstendiggjøring (planlagt deteksjon + onboarding-redesign):
  • Planlagt deteksjon (C1, opt-in): automatisk sjekk om Microsoft Learn-KB har endret seg. Tier 1 (SessionStart-hook kjører deteksjon når ≥ interval_days siden sist) + Tier 2 (lokal OS-scheduler launchd/cron, deteksjon-only). Claude-fri — poller sitemaps → rapport → discovery, oppdaterer ALDRI KB selv; operatør kjører /architect:kb-update manuelt.
  • Onboarding-redesign (C2): virksomhetskontekst + scheduler-config flyttet til bruker-eid ~/.claude/ms-ai-architect/ (overlever plugin-reinstall; lib/user-data.mjs ren resolver). Ambient injeksjon av kompakt org-sammendrag ved hver sesjonsstart (ikke bare status). Valgfritt fritekst-felt (free-context.md). Scheduler (enabled + os_scheduler_cadence) settes i onboarding via gated CLI (backup + atomisk skriving). Privat-sektor-paritet låst med regresjonsvakt (tests/test-onboarding-parity.sh): privat virksomhet fullfører onboarding uten «Annet» + privat reg-sett (DORA/Finansforetaksloven/Finanstilsynets IKT-forskrift/Verdipapirhandelloven/Hvitvaskingsloven).

Changed

• EU AI Act-tidslinje re-baselinet mot Digital Omnibus: Annex III høyrisiko (frittstående) utsatt til 2027-12-02, Annex I (innebygd) til 2028-08-02, Art. 50 transparens 2026-08-02. Art. 99-bøtesatser korrigert (35M €/7 %, 15M €/3 %, 7,5M €/1 %). Klargjort at EU AI Act ennå ikke er EØS-innlemmet og KI-loven uvedtatt per juni 2026.
• Agent-orkestrering currency: Connected Agents deprecated (pensjon 2027-03-31) → ny GA-modell Prompt/Hosted agents + Responses API som single entry point. MAF 1.0 GA + A2A v1.0.1. Prompt Flow (Foundry + AML) retirement 2027-04-20 → MAF. Computer Use Agent (CUA) GA 2026-05-07. Agentic retrieval GA-split (REST 2026-04-01 min/ekstraktiv, LLM-planning preview).
• Kostnadslag re-baselinet: GPT-5 $1,25/$10 input/output, konsolidert til én prissannhet i deterministic-cost-calculation-model.md.
• Foundry URL-navnerom-migrering: ai-foundry → foundry / foundry-classic på tvers av 141 filer.
• Dataresidens korrigert: Norway East — gpt-4o + gpt-4o-mini eneste Norge-residente generative modeller; modellkatalog modernisert.
• DPIA cross-border: EDPB Rec 01/2020 seks-stegs-TIA + CLOUD Act/FISA-restanalyse (FISA §702 til mars 2027; DPF gyldig; CLOUD Act uendret av DPF). EDPB 28/2024: anonymisering vurderes case-by-case.
• ROS + DPIA regulatorisk currency: NSM Grunnprinsipper v2.1, OWASP Agentic 2026 (ASI01–10), EchoLeak, DPF, EUDB.
• Sikkerhet: M365 E7 + Agent 365 (GA 2026-05-01) i licensing-matrix; Foundry Local air-gapped (Sovereign Private Cloud) i disconnected-scenarios.
• KB-routing: RAG/MLOps-KB ruting (betinget last) + forsont ROS-last-kontrakt (kjerne + betinget).
• Tellinger oppdatert: README docs-badge 387 → 389, README KB-tabell + per-skill-seksjon (ms-ai-security 60 → 62), CLAUDE.md skill-tabell (ms-ai-security 60 → 62), SKILL.md cost-optimization subdir-telling 21 → 22 (korrigert pre-eksisterende drift).

Notes on 1.16.0

• Verifisering: alle currency-påstander krysssjekket mot primærkilder. Der OWASP 2025-sider returnerte 404 ble innholdet kryssverifisert mot alternative offisielle kilder.
• Plugin API surface: 29 kommandoer, 12 agenter, 5 skills (389 reference docs), 2 hooks, MCP-server-config. Playground v3-koden er uendret siden v1.15.0 — kun doc-stamp bumpet; screenshots i playground/screenshots/v1.15.0/ ikke regenerert.
• Tester: 239 plugin-validering, kb-integrity 192/192 (220 baseline orphan-warns), kb-update 213, kb-eval 100, test-hooks 11/11, plugin-discovery 13/13, run-e2e alle suiter (security/cost/summary/ros/ai-act/onboarding-parity 14/14 + 6 playground-suiter) 0 FAIL. gitleaks: 3 pre-eksisterende i playground/vendor/.

[1.15.0] - 2026-05-16

Changed — playground v3 project-view integration

V2 project-surface (screen-tabs + category-tabs + per-command paste-cards) erstattet av v3 project-view (sidebar + main + import-modal). Sluttproduktet av 5 sesjoner: V2-PROJECT-VIEW-SPEC, JS-foundation, renderProjectView/renderProjectArtifact/renderArtifactNav, testing (97 nye assertions), og nå integration + ship.

Sesjon 5: integration + ship

• renderActive() ruter project-surface til ny renderProjectSurfaceV3() som wrapper renderProjectView + topbar + app-shell.
• renderProjectSurface (152 linjer), renderCommandSubCard (87 linjer), rehydratePasteImports (15 linjer) slettet.
• currentProjectScreen modul-var slettet; currentProjectTab beholdt som zombie for ACTIONS['project-tab']/ACTIONS['parse']/handlePasteImport (test-back-compat).
• ACTIONS['project-screen'] slettet.
• 5 v2-CSS-klasser slettet: .project-tabs, .project-tab, .project-tab__count, .sub-zone, .paste-import-row, .project-header__*, .command-cards.

Fingerprint-gap lukket

• requirements.headers: utvidet til /^\s*#\s*(EU\s*AI\s*Act\s*[—-]\s*Krav|AI\s*Act-?krav|Krav per|Requirements)/im (matcher "EU AI Act — Krav for høyrisiko provider+deployer").
• license.headers: utvidet til /^\s*#\s*(Lisens(kart|kapabilitets|-kapabilitets)?(legging|matrise)?|License\s*Mapping)/im (matcher "Lisens-kapabilitetsmatrise").
• KNOWN_GAP_FIXTURES = {} i tests/test-playground-fingerprints.sh (var { requirements: true, license: true }).

Migrasjon utvidet (v2→v3) med parserFor

• migrateDataVersion(state, archetypeFor, parserFor) tredje arg lagt til.
• Hvis reports[cid].parsed mangler men raw_markdown finnes, kjøres parserFor(cid) automatisk.
• defaultParserFor(cmdId) resolverer PARSERS[archetypeFor(cmdId)].
• Tre callsites oppdatert: cold-load, import-state, load-demo.

Browser-fixes funnet via dogfood

• components-tier4-project-view.css lagt til i <link>-chain (filen var vendored, men ikke loaded → modal-overlay og two-column layout virket ikke).
• renderImportModal setter data-open="true" på .import-modal-div (DS-kontrakt: display: flex aktiveres kun ved [data-open="true"]).

Tester

• bash tests/run-e2e.sh --playground → 386 PASS, 0 FAIL, 2 WARN (pre-eksisterende: .cmd-pipeline reservert; multiselect form-felt).
  • v3-static: 219 PASS (var 202 — la til 17 nye renderer-routing-asserts)
  • parsers: 70 PASS
  • migrations: 16 PASS
  • fingerprints: 32 PASS (var 30, 2 WARN → 32, 0 WARN)
  • project-view: 30 PASS
  • actions: 19 PASS
• bash tests/validate-plugin.sh → 219 PASS.

Screenshots regenerert til v1.15.0/

24 PNG-er (12 surfaces × 2 tema, retina, fullPage der applicable). Nye surfaces: project-overview, project-artifact-{classify, security, ros, cost, summary}, project-import-modal (viewport-only), project-search.

Demo-flyt verifisert i nettleser

• "Last inn demo-data" → 17 artifacts loaded + migrasjonen v2→v3 fyller parsed/verdict/keyStats.
• Sidebar viser alle 17 grupperte commands med severity-badges.
• Aggregate verdict (BLOKKERT) + key stats (17/17, 5/5, 2026-05-04) i header.
• Importer/Re-importer-modal åpner som overlay med backdrop.
• Per-artifact navigasjon (klikk i sidebar) → mounter riktig renderer i main-area.

Notes on 1.15.0

• Sesjon 5 fullført i én pass — token-budsjettet (~80-100k) holdt.
• v2-mockup.local.html + V2-PROJECT-VIEW-SPEC.local.md beholdt inntil sesjon 8 ship (per scope grenser).
• Pre-eksisterende kosmetiske issues (duplisert artifact-title, "Manglende rapporter"-heading-feil) ikke fikset — utenfor scope for sesjon 5 (integration), planlagt v1.16.0.

[1.14.0] - 2026-05-08

Changed — playground root-cause refaktor (6 sesjoner)

Etter v1.13.0 + v1.13.1 patchet 10+ symptomatiske visuelle bugs, dokumenterte v1.14.0-arbeidet at vi brukte DS via "band-aid lokal CSS" (191 linjer, 21 fix-kommentarer) i stedet for DS-konvensjoner. Brukeren valgte root-cause-fix etter å ha lastet ned Anthropic Plugin-Playground v3 referansepakke. Resultatet er DS-konvensjon-adopsjon på tvers av 14 renderere, lokal CSS halvert, og alle reported visuelle bugs løst på rotnivå.

Sesjon 2: DS root-cause fix (DS v0.3.x → v0.4.0)

Tre DS-bugs som tidligere ble symptomatisk patchet i lokal CSS er nå fikset i shared/playground-design-system/components-tier3-supplement.css og re-syncet til playground/vendor/:

• B-DS-1: .kanban-card__name. word-break: break-all knekket midt i ord — endret til word-break: break-word + overflow-wrap: anywhere. Tilsvarende lokal override i ms-ai-architect playground slettet.
• B-DS-2: .expansion__title-main + .expansion__title-sub. display: inline gjorde at title-main og title-sub kjørte sammen på samme linje. Endret til display: block så de stables vertikalt. Lokal override slettet.
• B-DS-3: .matrix__bubble. Manglet cursor: pointer, hover-effekt og :focus-visible outline når bobler er klikkbare. Lagt til som en del av base-mønsteret.

Sesjon 3: Risk-rapporter til DS-konvensjon

renderDpia, renderSecurity, renderRos refaktorert til DS-summary-grid + ros-layout-mønster fra Anthropic-ref ros-lier-kommune.html-scenario. Matrix i hovedcell + radar i <aside class="card">, top-risks som <ol class="top-risks"> inne i <div class="card">, findings med riktig list-only-mønster (BEM-struktur uendret). Defensive display:block; clear:both for .top-risks slettet — DS-pattern løser layout uten override.

Sesjon 4: Compliance/govern-rapporter til DS-konvensjon

6 renderere bytter lokal .report-meta-wrapper mot DS-mønstre:

• renderAiActPyramid — pyramide + tier-descs + <dl class="adr-meta"> for forpliktelser
• renderRequirements — <div class="findings"> med findings__list (master-detail riktig nå)
• renderConformity — kanban full bredde over, <section class="aiact-timeline-section"> full bredde under
• renderTransparency — ren <article>-struktur uten .report-meta-wrapper
• renderFria — DPIA-lignende mønster
• renderReview — kanban + suppressed-panel som tydelig under-seksjon

Sesjon 5: Phase-rapporter til expansion-list

renderMigrate + renderPoc bytter parallell-bullet .phase-detail-seksjoner mot <div class="expansion">-list (DS-supplement) — én expansion per fase med traffic-list inne i body. phases-summary-tabellen beholdt over expansion-listen for tabell-sammendrag. Lokal .phase-detail-CSS slettet.

Sesjon 5b: Lavt-scope renderer-fixes

Verifikasjon av tre renderere som ikke krevde DS-refaktor men hadde subtle bugs:

• renderCost key-stats. KEY_STATS_CONFIG['cost-distribution'] og inferVerdict('cost-distribution') viste "[object Object]" og returnerte alltid 'go' fordi parser-output har p50/p90 = {monthly, yearly}-objekter, ikke tall. Begge ekstraherer nå .monthly med fallback for flate fixtures.
• renderCompare vinn-attribusjon. firstWord-heuristikk feilet når begge subjekter delte førsteord (f.eks. "Azure AI Foundry" vs "Azure ML + AKS" ga begge fw='azure', kollapset vinn-attribusjon). Erstattet med distinctive-token-matching: full-subject-substring først, deretter ord som er unike for ett subjekt.
• renderUtredning aria-pattern. Droppet misvisende role="tab" / role="tablist" (vi rendrer ikke ekte tab-toggle, alle paneler er synlige). Beholdt aria-current="true" for visuell aktiv-markør. Ekte tab-toggle defer til v1.15.0.

Removed

• Lokal .phase-detail-CSS (overflødig etter expansion-list-refaktor)
• Lokal .report-meta-wrapper-CSS (alle 6 compliance/govern-renderere bytter til DS-pattern)
• Defensive containment-overrides (top-risks/aiact-timeline/phase-detail clear: both; display: block; width: 100%)
• Tre lokal-overrides slettet i sesjon 2: .kanban-card .kanban-card__name { word-break: break-word }, .expansion__title-main, .expansion__title-sub { display: block }, og matrix-bubble cursor/hover-overrides

Notes on 1.14.0

• Lokal <style>-blokk: 191 → 122 effektive linjer (~36% reduksjon siden v1.13.1)
• DS bumpet til v0.4.0 (delt mellom plugins, andre re-syncer på eget tempo)
• Alle 17 renderere PASS visuell QA mot demo-data i begge themes
• Plugin-validering: 219 PASS. Playground E2E: 272 PASS. Migrations: 7 PASS. Full E2E: 42 PASS.
• 24 screenshots (12 surfaces × 2 tema, retina, fullPage) regenerert til playground/screenshots/v1.14.0/. v1.10.0 + v1.11.0 beholdt som historisk referanse.
• Refs V1.14.0-PLAN.local.md + V1.14.0-AUDIT.local.md (slettet ved ship per plan).

[1.13.1] - 2026-05-06

Fixed — playground visuelle bugs (post-v1.13.0 manuell QA)

Patch-pakke som adresserer 10 visuelle bugs identifisert av maintainer i nettleser etter v1.13.0-shipped. Alle skyldtes mismatch mellom playground-rendrere og DS-konvensjoner som v1.13.0 ikke fanget opp.

• B7: classify "Forpliktelser" indent. <dl>+<ul> inne i <section class="report-meta"> brukte browser-default margin/padding som ga uforutsigbare indents. Lokal .report-meta-reset (DL som CSS-grid max-content + 1fr, h4 uppercase + bold, ul med padding-left: var(--space-5)).
• B8a: requirement-expand handler missing. renderRequirements markup hadde data-action="requirement-expand" på hver expansion__head-knapp, men ingen handler var registrert i ACTIONS-mappen. Klikk på R-01..R-09-radene i AI Act-krav-rapporten gjorde derfor ingenting. Fix: register ACTIONS['requirement-expand'] med samme toggle-mønster som onboarding-toggle-group.
• B8b: expansion title-main + title-sub kjørte sammen. DS-spans var inline (display: inline), så "dokumentertKilde: Art. 9" rendret uten linjebrytning. Lokal .expansion__title-main, .expansion__title-sub { display: block } så de stables vertikalt.
• B10: kanban-card tegnknekking. DS' .kanban-card__name har word-break: break-all som knekker midt i ord ("Tekn isk dokumen tasjon"). Lokal override .kanban-card .kanban-card__name { word-break: break-word }.
• B11: DPIA matrix-bobler ikke responderer på klikk. v1.13.0 click-handler matchet kun mot første-kolonne i Trusler-tabellen (T-001-mønster). Men DPIA-fixturer har full-tekst label i matrix_cells-data og T-001..T-005 i threats-tabellen — ingen match. Fix: utvid match-strategi til (Pass 1) exact first-cell, deretter (Pass 2) substring-match mot enhver celle med 40-tegn-prefiks-toleranse for å håndtere truncation.
• B12, B13, B15: defensive layout for top-risks/suppressed-panel/phase-detail/aiact-timeline. Eksplisitt display: block; clear: both; width: 100% på sibling-rapport-seksjoner som potensielt kunne lekkefra grid-elementer (.small-multiples, .kanban-board, .mat-ladder). .phase-detail får også margin-top + h3/h4-spacing-reset.
• B14: Migrate "skal vel være tabell". Lagt til phases-summary-tabell over phase-detail-seksjonene (Fase, Varighet, Milepæler-count, Suksesskriterier-count, Status). Samme tabell speilet i renderPoc for konsistens.

Notes on 1.13.1

• B9 (transparensnotis spacing) var visuell preferanse, ikke konkret bug — utsatt.
• Smoke-test 23/23 PASS (/private/tmp/claude-smoke-msarch-v1131.mjs).
• Plugin-validering 219 PASS, KB-update 42 PASS, playground E2E 271/271 PASS.
• Patch berører kun lokal CSS i <style>-blokk, ACTIONS-handler-registrering, click-handler-utvidelse, og to renderer-funksjoner. Ingen modifisering av playground/vendor/. Vendored DS' .kanban-card__name { word-break: break-all } står — overstyres lokalt.

[1.13.0] - 2026-05-06

Fixed — playground visuelle DS-bugs

Fix-pakke som speiler llm-security v7.6.1 (commit f9b555a). Samme klasse av visuelle bugs identifisert i ms-ai-architect playground via parallell DS-analyse. Verifisert i 22 smoke-test-asserts + full E2E-regresjon.

• B1: renderFindingsBlock + renderRequirements outer-wrapper. DS' .findings er display: grid; grid-template-columns: 360px 1fr (list+detail-panel). Playground bruker bare list-kolonnen, så outer <div class="findings"> klemte indre struktur til 360px og forlot 1fr-kolonnen tom. Erstattet med <section class="report-meta">-wrapper. Indre findings__list > findings__group > findings__group-header + findings__items BEM-struktur uendret.
• B2: .report-table lokal CSS. Brukt i 6+ rapporter (Trusler, Kostnadsoversikt, TCO, Risiko-tabell, Key Metrics) men ikke definert i vendored DS — rendret som ren ustylet tekst-grid. Lokal styling speilet fra llm-security v7.6.1: 100%-bredde, border-collapse, bg-soft th-bakgrunn med uppercase-overskrift, hover-rad, mono-formatert <code>-inline.
• B3: ROS-matrise klikkbare bobler. <span class="matrix__bubble"> byttet til <button type="button" data-threat-id="…" aria-label="Trussel: …">. Document-level click-handler matcher .matrix__bubble[data-threat-id], scroller smooth til tilsvarende rad i Trusler-tabellen og highlighter raden i 1.6 sek. Lokal CSS for cursor:pointer, transform: scale(1.15) på hover, :focus-visible outline.
• B4: renderRadarSvg label-overlap ved 6+ akser. Bumpe SVG fra 300×300 til 380×380, R fra 100 til 125, label-offset fra R+25 til R+28. Dynamisk text-anchor basert på horisontal-posisjon (Math.abs(dx) < 0.2 ? 'middle' : (dx > 0 ? 'start' : 'end')) for å unngå at bottom-labels overlapper hverandre — typisk for ROS-rapporten med 7 risiko-dimensjoner.
• B5: recommendation-card__body overflow-wrap. Lange single-line tekster (URLer, owner-tags, datoer) skubbet ut av viewport i grid-cellen auto + 1fr. Lokal CSS overflow-wrap: anywhere; word-break: break-word;.

Changed

• tests/test-playground-v3.sh: DS-klasse-assertion oppdatert fra .findings til .findings__list (BEM-list er fortsatt i bruk; outer grid-container bevisst fjernet i B1).

Notes on 1.13.0

• Bugfix-only release. Ingen scanner-/agent-/knowledge-endringer. Ingen modifisering av playground/vendor/ (DS-endringer der må gå via shared/playground-design-system/ + re-sync).
• Playground v3-arkitektur uendret. Alle 17 rapport-renderers og parser-routing-tabell uendret.
• Smoke-test: 22/22 PASS (/private/tmp/claude-smoke-msarch-v1130.mjs). E2E playground: 272/272 PASS (201 statisk-struktur etter test-oppdatering + 70 parser-fixtures + 1 verdict-pill). Plugin-validering: 219 PASS. KB-update: 42 PASS.

[1.12.0] - 2026-05-05

Added — Manuell KB-refresh-arbeidsflyt

• commands/kb-update.md — ny /architect:kb-update slash-kommando som driver hele KB-oppdaterings-løypen fra en aktiv Claude Code-sesjon: run-weekly-update.mjs --force --discover → report-changes.mjs → per-fil microsoft_docs_fetch → Edit-baserte oppdateringer → git commit. Argumenter: --skip-discover, --priorities <list>, --dry-run, --single-commit. Default-prioritet: critical,high. Schedulering er eksplisitt utenfor scope og overlatt til brukeren.

Removed — launchd/cron-arkitektur (Wave 3-5 reversert)

Etter ToS-gjennomgang (Anthropic Consumer Terms § 3 — automated access only via API key or where explicitly permitted) ble den autonome cron-arkitekturen vurdert som unødvendig kompleks for en solo-fork-and-own-plugin. Apply-fasen krever LLM-resonnering uansett, så manuell trigger fra en aktiv Claude Code-sesjon er både enklere og holder pluginen klart innenfor ToS.

• scripts/install-kb-cron.mjs (~400 linjer) — cross-OS cron-installer (launchd/systemd/Windows Task Scheduler)
• scripts/kb-update/weekly-kb-cron.mjs (~600 linjer) — cron-orkestrator med pre-flight, lock, backup, claude -p subprocess, post-run verify, rollback
• scripts/kb-update/templates/ — 4 scheduler-templates (com.fromaitochitta.ms-ai-architect.kb-update.plist, ms-ai-architect-kb-update.{service,timer,ps1}) + README
• scripts/kb-update/lib/auth-mode.mjs (~100 linjer) — detectAuthMode + validateAuthForCron (kun cron-spesifikk validering)
• scripts/kb-update/lib/lock-file.mjs (~120 linjer) — PID+mtime stale-detection (kun for cron-kollisjons-vern)
• scripts/kb-update/lib/cost-estimat.mjs (~80 linjer) — pre-flight budget-cap-logikk (kun for api-key cron-kjøringer)
• tests/kb-update/test-{auth-mode,lock-file,cost-estimat,install-cron,session-start-status,template-generation,weekly-kb-cron-flags}.test.mjs — 7 testmoduler for slettet kode (~68 testtilfeller)
• tests/test-kb-update.sh — Wave 5 Bash-3.2-shim, erstattet av direkte node --test-kall i run-e2e.sh
• hooks/scripts/session-start-context.mjs: kb-update-status.json-overvåking (Wave 3 Step 10) + ubrukt getCacheDir-import fjernet

Total: ~1500+ linjer kode + 7 testmoduler ut. Beholdte utilities (atomic-write, backup, cross-platform-paths, log-rotate) er fortsatt brukbare av run-weekly-update-scriptet og kan gjenbrukes i fremtidige skripts.

Changed

• tests/run-e2e.sh --kb-update kaller nå node --test tests/kb-update/*.test.mjs direkte (4 testmoduler, 42 tester) i stedet for bash tests/test-kb-update.sh-shim
• README.md — "Knowledge Base Maintenance"-seksjonen rewriter for manuell modell, scheduling-disclaimer lagt til
• CLAUDE.md — KB-ferskhet-seksjonen oppdatert, /architect:kb-update lagt til i kommando-tabellen

Notes on 1.12.0

• ToS-vurdering: kjøring fra aktiv Claude Code-sesjon faller innenfor "Claude Code CLI is exempted from the prohibition on automated access" per Anthropic auth-docs og Consumer Terms § 3. Apply-fasen kan ikke automatiseres innenfor pluginens scope — eventuell schedulering er forkers eget ansvar
• Forbruk: én manuell kjøring (default --priorities critical,high) henter ~80 Microsoft Learn-sider via microsoft_docs_fetch og oppdaterer 9-53 filer. Kvote-bruk avhenger av abonnementets rate-limits — typisk innenfor en daglig Pro/Max-allowance
• 42/42 KB-update-utility-tester PASS. 271/271 playground-tester uendret
• data/change-report.json blir værende på disk for diagnose etter hver kjøring

[1.11.0] - 2026-05-04

Added — Design-system 100%-adoption + visual upgrade

The playground now consumes the shared playground-design-system v0.3.0 end-to-end. Plugin contract (24 commands, 12 agents, 5 skills, 2 hooks, MCP) is unchanged — playground-internal release.

• Hoisted to shared DS (v0.3.0) — 13 generic components moved from inline <style> to shared/playground-design-system/components-tier3-supplement.css: .eyebrow, .page__*, .key-stats/.key-stat, .verdict-pill-lg (5-band), .tab-list/.tab/.tab-panel, .top-risks/.top-risk[data-severity], .recommendation-card, .card__* subcomponents, .card--severity-{level}, form-patterns (.field-row, .field-label, .field-help, .multi-select, .checkbox-row, .required-mark), .stack-lg/-md/-sm, .pyramide-tier-detail, .scenario-card-grid/.scenario-card[data-status="winner"], .app-shell variants. 567 new lines, 107 new selectors. Re-synced to vendored DS via scripts/sync-design-system.mjs ms-ai-architect.
• PARALLEL CSS migration — all 6 plugin-local naming groups renamed to DS conventions: .topbar* → .app-header*, .residual-pair* → .pair-before-after*, .command-card* / .catalog-card* → .card + .card__*, .screen-tabs/-tab/screen → .tab-list/.tab/.tab-panel, .pyramide-desc* → .pyramide-tier-detail*. Includes markup restructure for tab-component (data-active="..." → [hidden]-attr) and pair-before-after (data-severity → BEM modifier). ~50 markup occurrences updated.
• Inline <style> trim — block reduced 202 → 127 lines (~37%). renderPageShell + renderKeyStatsGrid refactored to DS markup (.page__header-main + .page__header-aside + h1.page__title; .key-stat--{level} BEM modifier).
• Visual upgrade (benchmarked against showcase) — AI Act pyramid width fix (min-width 480px, font-size-md, responsive @media), severity-coded left-border on report-cards in project surface based on parsed.verdict, app-header restructure (brand + breadcrumb + spacer + actions), eyebrow-label "PROSJEKTER · X av X" on home, .stack-lg vertical rhythm on home/project/catalog body.
• Demo state rename — Acme AS → Acme Kommune and Demosystem → Acme Kunde-chatbot consistently across all 17 fixtures (39 Acme occurrences). build-demo-state.mjs updated: organization.name = "Acme Kommune", projects[0].id = "acme-kunde-chatbot", projects[0].name = "Acme: Kunde-chatbot".
• playground/screenshots/v1.11.0/ — 24 regenerated PNGs (~27 MB total) reflecting the visual upgrade. v1.10.0 set preserved as historical reference.
• tests/screenshot/run.mjs — OUT_DIR + comments bumped to v1.11.0.

Plugin-local survivors (intentional)

Components kept inline because they are semantically distinct from DS counterparts:

• .verdict-pill (domain go/block semantics, distinct from DS .verdict-pill-lg severity bands)
• .scenario-card[data-status="met/partial/missing"] (DS only ships "winner")
• .read-more-block + .suppressed-panel (native <details>; DS uses JS-toggled aria-expanded)
• Surface-specific layouts: .onboarding-*, .home-*, .project-*, .modal*, .command-form*, .catalog-cards-container

Notes on 1.11.0

• 3-session rollout (commits 40631c0 Fase 1-3 + 31aed40 Fase 4-6 + 7ffaa82 Fase 7-9). Session 1+2 committed with [skip-docs] because user-facing docs are bundled with the release commit.
• Tests: 201 static + 70 parser-fixture + 7 migration = 278 PASS playground-suite, 0 FAIL. Pre-existing tests/test-ros-output.sh fixture-missing failure is unchanged.
• DS bump is purely additive — other vendor consumers (okr, llm-security, ultraplan-local, config-audit) do not need to re-sync to remain green.

---

[1.10.1] - 2026-05-04

Added — Demo system + screenshot gallery

The playground now ships with a one-click demo and committed screenshots so forkers can see what the plugin produces without running anything.

• <script type="application/json" id="demo-state-v1"> inline block — built by scripts/build-demo-state.mjs from playground/test-fixtures/*.md. Contains a complete state: 18 onboarding fields prefilled for "Demo kommune", one demo project with all 17 reports as raw_markdown. ~37 KB inline, file://-safe (no fetch).
• "Last inn demo-data" button on the onboarding action-bar. Clicking it replaces all state with the demo and navigates to the project surface.
• raw_markdown persistence — handlePasteImport now writes markdown to project.reports[id].raw_markdown (with equal-value guard to prevent render loops). Survives page reload.
• rehydratePasteImports() — on project surface mount, fills textareas from raw_markdown and re-renders visualizations. Hooked via queueMicrotask after innerHTML commit.
• tests/screenshot/ tooling — standalone Playwright runner (its own package.json, gitignored node_modules). Captures 24 PNGs per release (12 surfaces × 2 themes) at deviceScaleFactor: 2 retina, fullPage: true.
• playground/screenshots/v1.10.0/ — 24 committed screenshots (~28 MB total) covering onboarding, home, catalog, project surface (rapporter × 5 tabs, oversikt, kontekst, eksport).

Notes on 1.10.1

• v1.10.0 was tagged earlier on 2026-05-04 with the foundation refactor. v1.10.1 ships the demo + screenshots without changing any v1.10.0 behavior — purely additive: one inline JSON block, one button, two new functions, one tooling directory.
• Plugin contract (24 commands, 12 agents, 5 skills, 2 hooks, MCP) remains unchanged.
• Tests: 215 plugin-validation + 201 + 70 + 7 = 493 PASS. Pre-existing tests/test-ros-output.sh fixture-missing failure is unchanged.

---

[1.10.0] - 2026-05-04

Added — Playground v3 felles grunnskjelett (foundation refactor)

All 17 report renderers now share a unified visual signature through renderPageShell(). The plugin contract (24 commands, 12 agents, 5 skills, 2 hooks, MCP) is unchanged — this release is playground-internal.

• Foundation helpers: renderPageShell(), renderVerdictPill(), renderKeyStatsGrid(), inferVerdict(), inferKeyStats(), KEY_STATS_CONFIG — provide eyebrow + h1 + optional 5-band verdict-pill + optional 3-5 key-stats-grid that every renderer wraps around its archetype-specific body.
• Schema v1→v2 migration: Eager MIGRATIONS pipeline at cold-load and import time, idempotent via dataVersion=2 guard. Backfills verdict and keyStats on existing reports without disrupting user state. tests/test-playground-migrations.sh (7 PASS) validates idempotency.
• Tier 3 supplement components integrated: kanban (conformity, review), mat-ladder (migrate, poc), screen-tabs (utredning, project surface), scenario-card-grid (license, compare), residual-pair (dpia, ros), top-risks (ros), recommendation-card (security, ros), suppressed-panel (review), critique-card (adr), read-more (utredning, summary), traffic-light (poc).
• Parser extensions (R15 forward-compat):
  • parsePhasedPlan — phases[].status (planned/active/done) + currentPhaseIndex + pocVerdict
  • parseComparison — winner field from ## Vinner: <id>-line
  • parseMatrixRisk — _consumer discriminator (Strategy A) for shared dpia + ros usage
• Onboarding redesign: Exactly 4 structured fields (select / multiSelect for sector, regulations, license, scenarios) + 14 free-text fields with placeholder hints. Improves prefill flexibility while keeping enum-driven fields where they matter.

Added — Light theme tokens

• [data-theme="light"]-block in shared/playground-design-system/tokens.css (re-synced to playground/vendor/) with verbatim Aksel values. Both themes meet WCAG 2.2 AA contrast.
• Theme-bootstrap-script in <head> reads saved localStorage('ms-ai-architect-theme') before stylesheets parse — no flash-of-wrong-theme.

Added — v1.10.0 fixtures and validation

• tests/test-playground-migrations.sh — new 7 PASS suite for v1→v2 idempotency.
• tests/fixtures/playground/state-v1-snapshot.json — pre-migration state fixture.
• tests/test-playground-v3.sh extended to 201 PASS — new asserts on per-renderer renderPageShell usage, Tier 3-bruk (kanban / mat-ladder / screen-tabs / scenario-cards), onboarding field-distribution (4 strukturerte / 14 fritekst), _consumer-strategi A.
• Regenerated fixtures: migrate.md, poc.md (per-phase Status + POC-Verdict), compare.md (## Vinner:-line).
• playground/A11Y-RAPPORT.md — populated with code-based static assessment of all 4 surfaces and 17 renderers (was: skeleton). Browser-axe-core run still pending per MANUAL-CHECKLIST.md section 10.

Changed

• playground/ms-ai-architect-playground.html — all 17 renderers wrap output via renderPageShell(). Inline CSS additions (e.g. .pyramide-desc, .scenario-card-grid, .residual-pair, .read-more-block, .top-risks, .recommendation-card, .suppressed-panel) are flagged as v1.11.0 candidates for hoisting to shared/playground-design-system/components-tier3-supplement.css.
• Plugin README — refreshed v1.10.0 section with foundation-helpers, Tier 3 adoption list, parser extensions, schema migration, theme-tokens.
• Plugin CLAUDE.md — Playground (v3) section now reflects v1.10.0 architecture, validation counts (201 + 70 + 7), and inline-CSS hoisting candidates.
• Marketplace root README — bumped ms-ai-architect to v1.10.0 with felles grunnskjelett summary.

Notes

• Plugin API surface unchanged. All 24 commands, 12 agents, 5 skills (387 reference docs), 2 hooks, and MCP server configuration remain identical to v1.9.0. v1.10.0 is a minor bump per SemVer because the playground state schema change is auto-migrated and backward-compatible.
• 6-session implementation per .claude/projects/2026-05-03-ms-ai-architect-v1-10-playground/plan.md:
  • Session 1: Foundation (theme tokens, helpers, MIGRATIONS, v1-snapshot fixture)
  • Session 2: Onboarding redesign + surface page-headers + screen-tabs
  • Session 3: Sub-batch A renderers (regulatory 6) — aiact-pyramid, requirements, transparency, frimpact, conformity, dpia
  • Session 4: Sub-batch B renderers (security 3) — security, ros, review (REFERENCE STANDARD)
  • Session 5: Sub-batch C renderers (econ + dokumentasjon 8) — cost, license, migrate, adr, summary, poc, utredning, compare
  • Session 6: Visual QA + A11Y-rapport + dokumentasjon-overhaul + version bump (this commit)
• Pre-existing test pollution: tests/run-e2e.sh references tests/test-ros-output.sh which expects a fixture that has never existed in git (tests/fixtures/ros-analysis/fixture.md). Reports as 1 suite failure but is unrelated to v1.10.0 — predates this work and is unchanged here.
• Session 4 commit 645f016 (B.1 bundling): That commit message was bundled with ultraplan-local work due to hook timing — historical only, not relevant to plugin behavior.

---

[1.9.0] - 2026-05-03

Added — Playground v3 (full architecture rewrite)

The playground at playground/ms-ai-architect-playground.html was rebuilt from scratch as a multi-surface decision-builder + report viewer (3867 lines, single-file). It replaces the v2 5-step pipeline with a project-aware app that persists state across sessions and visualizes parsed reports inline.

• 4 surfaces:
  • Onboarding — 5 grouped expansion sections (organization, technology, security, architecture, business) with 18 fields covering organization profile, tech stack, compliance, architecture preferences, and business references. Drives prefill across all 24 commands.
  • Home — 3 entry tracks (Onboard / New project / Catalog) plus project list with completion-meter per project.
  • Catalog — 24 commands grouped in 5 expansion categories (regulatory, security, economy, documentation, tool) with full-text search, filtered counts, and direct form-launch.
  • Project — per-project tabs (one per command category), command form prefilled from shared onboarding state, paste-back markdown import that parses and visualizes reports inline.
• Persistent state: IndexedDB primary store with localStorage fallback. Schema-versioned (STATE_KEY = 'ms-ai-architect-state-v1') with eager MIGRATIONS pipeline that runs at cold-load and import time. Throttled writer prevents excessive disk traffic.
• 17 inline report renderers routed via canonical archetype-routing table:
  • Regulatory (6): classify pyramid, requirements list, transparency text, FRIA, conformity checklist, DPIA matrix-risk-6×5
  • Security (3): security 5×5 matrix, ROS 5×5 matrix with 7-dimension threat library, review findings
  • Economy (2): cost distribution P10/P50/P90, license capability matrix
  • Documentation (6): migrate phased plan, ADR, summary markdown, POC verdict, utredning, compare
• 14 markdown parsers with the same archetype keys, exposed as window.__PARSERS. All parsers return {ok, data} or {ok, errors[]} — never throw.
• Light/dark theme toggle in the topbar, persisted in localStorage('ms-ai-architect-theme'). FOUC-safe: a small bootstrap script in <head> reads the saved value before stylesheets parse.
• Export/import as JSON Decision Record envelope (Blob + FileReader). Schema-version-aware on import; downgrades trigger MIGRATIONS automatically.
• Vendored design-system at playground/vendor/playground-design-system/, kept in sync via scripts/sync-design-system.mjs ms-ai-architect from the marketplace root. SHA-256 MANIFEST detects local drift.
• 24 ACTIONS registered through a single delegated click handler on document — keeps event-handler footprint minimal.

Added — Playground validation

• tests/test-playground-v3.sh — bash 3.2-compatible static structure tests (170 PASS lines): vendored CSS link order, file://-safety, surfaces, 24 commands, 14 parsers, 17 renderers, design-system class usage, exposed globals, 23 ACTIONS handlers.
• tests/test-playground-parsers.sh — 70 PASS lines: validates 17 fixtures × parser routing keys + handlePasteImport wiring.
• tests/run-e2e.sh --playground — new flag dispatching both suites; included in --all.
• playground/test-fixtures/ — 17 synthetic markdown fixtures (one per report-producing command) covering all canonical parser archetypes. Generic example data only.
• playground/A11Y-RAPPORT.md — accessibility report skeleton with WCAG 2.2 AA test grid for both themes.
• playground/MANUAL-CHECKLIST.md — 10-section manual QA checklist (round-trip, schema migration, project CRUD, command form prefill, paste-import per report type, parse errors, export/import cycle, theme toggle, file://-standalone, axe-core a11y per surface) with concrete DevTools-console assertions.

Changed

• playground/ms-ai-architect-playground.html — replaced (v2 1990 lines → v3 3867 lines). Same canonical filename, so external links continue to resolve.
• Plugin README — new "## Playground (v3)" section documenting the 4-surface architecture and validation matrix.
• Plugin CLAUDE.md — replaced v2 5-step pipeline section with v3 architecture overview. docs/playground-v2-spec.md retained as historical reference but no longer the contract; v3 spec lives at .claude/projects/2026-05-03-playground-v3-architecture/.
• Marketplace root README — playground listing for ms-ai-architect updated to describe v3 architecture, persistence model, 17 renderers, theme toggle, and 240-test validation.

Notes

• Plugin API surface unchanged. All 24 commands, 12 agents, 5 skills (387 reference docs), 2 hooks, and MCP server configuration remain identical to v1.8.0. Playground v3 is an additive, optional UI surface — users invoking /architect:* directly see no behavioral difference.
• v1.9.0 is a minor bump per SemVer because the plugin contract (commands/agents/skills/hooks/MCP) is backward-compatible. The playground's local IndexedDB state schema would only affect users who had the v2 playground running; existing state is auto-migrated by the eager MIGRATIONS pipeline.

---

[1.8.0] - 2026-04-09

Added

• Sitemap-based KB change detection system (scripts/kb-update/)
  • build-registry.mjs — extracts 1342 unique URLs from 387 reference files
  • poll-sitemaps.mjs — streams 18 Microsoft Learn sitemaps for <lastmod> changes
  • report-changes.mjs — prioritized change report (critical/high/medium/low)
  • discover-new-urls.mjs — finds relevant new MS Learn pages not yet covered
  • run-weekly-update.mjs — weekly orchestrator with --force, --discover, --dry-run
• Shared libraries: url-normalize.mjs, sitemap-stream.mjs, registry-io.mjs (zero npm deps)
• Background poll trigger in session-start hook (auto-runs if >7 days since last check)

Changed

• Session-start hook reads sitemap-based change report instead of broken mtime staleness check
• generate-skills --update reads change-report.json for targeted MCP updates with changed_urls
• KB maintenance section in README updated with automated freshness and auto-update info

Deprecated

• scripts/kb-staleness-check.sh (mtime-based, unreliable after git clone — kept for backwards compat)

---

[1.7.0] - 2026-04-07

Changed

• Open-source release in ktg-plugin-marketplace
• Moved skill-gen tooling from orchestrator/ to scripts/skill-gen/
• Fixed broken DPIA knowledge base reference path (norwegian-public-sector-governance/ → responsible-ai/)
• Removed orphaned pre-edit-secrets hook (secrets scanning consolidated to llm-security plugin)
• Updated all badge counts (version 1.7.0, 387 ref docs, 12 agents)
• Fixed agent model column in CLAUDE.md (sonnet → opus, matching actual frontmatter)
• Added missing ai-act-assessor agent to README agent table
• Added Bash scope guardrail to commands/export.md
• Added supply chain warning for remote install scripts in knowledge base

Removed

• Internal orchestrator tooling (Linear integration, hardcoded issue IDs)
• Client-specific utredning examples (docs/utredning/)
• Generated test report artifacts (e2e-test-report.md, .pdf)

---

[1.6.0] - 2026-02-19

Added

• ROS analysis command and agent (/architect:ros)
• 7-dimension risk assessment with NS 5814/ISO 31000 methodology
• AI threat library (49 threats across 7 categories)
• Sector-specific checklists (health, transport, finance, justice, education)
• MAESTRO multi-agent security model integration
• 7 new KB reference documents (3,131 lines)
• E2E test suite for ROS (24 checks)
• summary-agent integration for ROS output

---

[1.5.0] - 2026-02-13

Added

• E2E regression tests (43 checks across 3 suites: security, cost, summary)
• Auto onboarding detection at session start
• Systematic KB update process with staleness policy
• kb-staleness-check.sh with --json output mode

---

[1.4.0] - 2026-02-13

Added

• Onboarding agent (/architect:onboard) — 5-phase structured interview
• README rewrite to English

---

[1.3.0] - 2026-02-13

Changed

• 5-skill migration: 1 monolithic skill → 5 domain-specific skills (ms-ai-advisor, ms-ai-engineering, ms-ai-governance, ms-ai-security, ms-ai-infrastructure) with 364 reference documents
• Fixed 13 broken KB references
• Fixed UTF-8 encoding issues across reference files

---

[1.2.0] - 2026-02-13

Added

• Runtime hooks: secrets detection (pre-edit-secrets.mjs), session context, stop reminders
• Test infrastructure: hook tests, KB integrity tests, plugin discovery tests
• /architect:export command for PDF export

---

[1.1.0] - 2026-02-13

Added

• summary-agent: executive summary and decision memo synthesis
• dpia-agent: DPIA/PVK for AI systems
• Utredning orchestrator v2
• Production readiness (21 bug fixes)

---

[1.0.0] - 2026-02-12

Added

• Initial release: 20 knowledge bases, 8 agents, architecture-review-agent
• Cosmo Skyberg architect persona with 7-phase advisory methodology
• MCP integration with Microsoft Learn for live documentation verification
• Core commands: /architect, /architect:research, /architect:security, /architect:cost, /architect:adr, /architect:diagram, /architect:review, /architect:compare

---

[0.4.0] - 2026-02-07

Note

First formal version. Previously unversioned.

What existed

• 145+ reference documents across 14 categories
• 6 specialized agents: research, security assessment, cost estimation, ADR writer, license mapper, diagram generation
• 13 commands: research, compare, cost, license, security, poc, adr, migrate, utredning, generate-skills, diagram, setup, help
• Cosmo Skyberg architect persona
• MCP integration with Microsoft Learn
• Session logging with ADR reminders