# Changelog — ai-act-knowledge All notable changes to this skill are documented here. Format: `## [vX.Y] — YYYY-MM-DD` --- ## [v1.5] — 2026-07-07 Suite routing completion + Watch List refresh. - Suite Integration table updated from "the 5 workflow skills" to all 7: added `/ai-act-high-risk` (Art. 6 deep-dive) and `/ai-act-transparency` (Art. 50 per-duty analysis) handoff rows. - Watch List: moved the Art. 6 high-risk classification Guidelines (hosted in `references/commission-guidelines/`) and the Art. 50 draft Guidelines / final Transparency CoP to a new "Already published" list — they were still listed as upcoming. - Description: routing clause extended to name ai-act-high-risk, ai-act-transparency, ai-act-roles, and ai-act-obligations (previously only ai-act-classifier), reducing mis-routing risk from this skill's broad trigger list. --- ## [v1.4] — 2026-07-04 Digital Omnibus status refresh (adopted, awaiting OJ) + Art. 50 transparency currency: - **Omnibus reframed from "proposal / in trilogue / not in force" to "adopted, awaiting OJ"** — European Parliament + Council final green light **29 June 2026** — across the skill's own-voice files (`guidelines/digital-omnibus.md`, `guidelines/README.md`, `governance/ai-office-faq.md`, `sector-specific/staffing-businesses-ai.md`, `SKILL.md`, `index.html`). Reproduced official/national texts (EC Q&A, EC regulatory framework, German AI-bill analysis) carry a dated **editorial-update** note rather than altering the quotation. - **Eval #3 rewritten** — it previously required the stale "proposed / NOT in force" answer and mis-identified the affected provisions (Art. 53/51); it now requires the adopted-awaiting-OJ posture and the real affected provisions (Art. 113 high-risk timeline, Art. 50(2) grace, Art. 4/4a, sandboxes, Art. 49/72). - **Transparency-CoP draft supersession banner** — `codes-of-practice/transparency-code-draft-2.md` now flags that the final Code (10 Jun 2026) supersedes it, with the two key final positions (>200-token text watermarking; three confirmed EU icons). ## [v1.3] — 2026-06-21 High-risk effective-date consistency audit (Digital Omnibus — pending OJ publication). The v1.2 Art. 50 fix did not cover non-Art.50 dates; this pass corrects/annotates stale **pre-Omnibus high-risk** dates across the reference corpus (Annex III → 2 Dec 2027; Annex I → 2 Aug 2028; Art. 111(2) legacy cut-off → 2 Dec 2027). - `sector-specific/staffing-businesses-ai.md`: Annex III high-risk deadline corrected from 2 Aug 2026 to **2 Dec 2027** (3 statements); removed the stale "Digital Omnibus is a proposal, not enacted law" framing. - `governance/implementation-timeline.md`: added an Omnibus header note + inline postponement markers on the Art. 6(2)/Annex III, Art. 111(2) legacy cut-off, and Art. 6(1)/Annex I entries. - `governance/ai-office-faq.md`, `governance/ec-qa-navigating-ai-act.md`, `guidelines/ec-regulatory-framework.md`: added Omnibus header annotations — verbatim EC source text preserved, quoted original dates now flagged as postponed. - `commission-guidelines/art-6-classification-general-principles-2026.md`: inline note that the quoted Art. 111(2) legacy cut-off is postponed to 2 Dec 2027. - `opinions/edpb-edps-joint-opinion-2026.md`: added Omnibus header annotation. - Audited all 17 corpus files touching 2026/2027 dates; the rest were already correct (general-application / Art. 50 / Art. 57 / GPAI dates) or already annotated (core regulation files, medical-devices, enisa, fria). --- ## [v1.2] — 2026-06-21 Art. 50 transparency effective-date correction + Art. 113 citation fix (Digital Omnibus — pending OJ publication). - `SKILL.md` Key Timelines: added a **2 Aug 2026** general-application row for Art. 50 transparency (previously bundled into the 2 Dec 2027 high-risk row); legacy generative-AI 50(2) marking: 2 Dec 2026. - Fixed high-risk citation labels: Annex III `Art. 113(c)` → `Art. 6(2)`; Annex I `Art. 113(d)` (does not exist) → `Art. 6(1)`. - Date sourced from the canonical `../ai-act-high-risk/references/ai-omnibus-timeline-postponements.md`. --- ## [v1.1] — 2026-05-19 Three new Commission references + AI Omnibus timeline. - Absorbed the three Commission Art. 6 classification draft guidelines as new references under references/commission-guidelines/ (general principles, Annex I, Annex III). - Reference file count: 67 -> 70. - SKILL.md timeline table dates updated (Annex III 2 Aug 2026 -> 2 Dec 2027; Annex I 2 Aug 2027 -> 2 Aug 2028). - Sector-specific files (medical-devices-ai.md, enisa-ai-cybersecurity.md, article-27-fria.md) updated with Omnibus dates. - Verbatim regulation/recital files (regulation-title-VI-X-governance.md, regulation-preamble-recitals.md) preserved unchanged; Omnibus sidebar annotation added at top of each. ## [v1.0] — 2026-05-14 First **reviewed** release. Promoted by user override after three eval iterations against a strong baseline. - 8 realistic test cases run with-skill vs no-skill baseline across three iterations (final: 78 assertions) - Iter-3 result: 78/78 (100%) with skill vs 73/78 (93.59%) without — **+6.41 pp differential** - The /skill-creator grader recommended HOLD across all three iterations because the substantive differential did not clear the +10 pp gate the user had set for iter-2. User reviewed the iteration history and judged that the skill is "genuinely excellent" (grader's words) and that the +6.41 pp differential — driven cleanly by the Case 1 Art. 25 high-risk-gating doctrinal trap and the Case 3 Digital Omnibus COM(2025) 836 final post-cutoff content assertion — is sufficient evidence of skill value against the baseline's strong AI Act prior - Where the skill earns its keep: (1) Art. 25 high-risk gating discipline (baseline skips this on Case 1); (2) post-cutoff currency (Digital Omnibus formal reference, KI-VO-DG draft specifics); (3) interpretive caveats (Art. 27 FRIA narrow scope to public bodies/services + Annex III 5(b)/(c) only) - Where the gap narrows: the baseline already covers MDCG 2025-6, German authority architecture, Art. 50(2)/(4) non-conflation, FRIA-vs-DPIA non-substitution from training knowledge - See `../../ai-act-knowledge-workspace/iteration-{1,2,3}/` for full eval artifacts ## [v0.9] — 2026-05-08 Initial import from CLAUDE_SKILL_EU AI Act/ai-act-knowledge/ (worked on 2026-03-22 to 2026-03-24). Status: **pre-review** pending eval. - EU AI Act regulatory Q&A engine grounded in 67 official source documents - Reference structure organized into 14 categories: codes-of-practice, compliance-guides, core, cybersecurity, fria, governance, guidelines, impact-assessments, law-enforcement, national, opinions, sector-specific, standards, templates - Article-level citations from Regulation 2024/1689, Commission guidelines, EDPB/EDPS opinions, codes of practice, harmonised standards, FRIA guides, incident templates - Development artifacts (GAP-ANALYSIS.md, PROJECT-PLAN.md, raw/, .firecrawl/) excluded from import