# v0.9.9 → v0.9.10 — AIGC Content Creation Category + Research Expert Team Skill (Oct 13, 2026, part 3) ## Summary Added the **26th category — AIGC Content Creation** (22 methods) and the **18th skill — `research-expert-team`** (a nine-role full-cycle research expert team). Library: **934 → 955**; taxonomy: **25 → 26**; skills: **17 → 18**. All carriers re-verified at 955 with identical name sets. ## New Category: AIGC Content Creation (22 methods) **Positioning** (deliberately non-overlapping): the creation-side application layer of AI-generated content over 3DGS — text/image → 3DGS scenes and assets, generative scene reconstruction, stylization, artistic fonts, avatars. Generic 3D generative models stay in `Generation`, editing pipelines in `Editing`, watermark/copyright protection in `Security`. **Admission pipeline** (same as v0.9.9): arXiv API multi-query search (`.temp/search_aigc.py`, 12 queries, 135 unique candidates) → GS-relevance filter + de-dup vs. existing library (67) → shortlist R9 verification via `id_list` full-abstract fetch (`.temp/verify_aigc_shortlist.py`) → admission (`.temp/admit_aigc.py`, **22/22 R9 PASS: method names self-stated in title/abstract; all arXiv IDs reachable**) → CSV + abstracts.js append → `build_knowledge_base.py` regeneration. - 2026 entries (7): VirSqueezer, CGGS, UniWorld-View, OmniStyle-INR, Flex4DHuman, LightHarmony3D, PSGS - 2025 anchors (15): SplatFont3D, InStyle, DirectGaussian, Drag4D, Lyra, T2Bs, DreamArt, CSD, GaussVideoDreamer, GS-RGBN, DD3G, GSV3D, Dragen3D, GaussianMotion, DiffSplat - **2 candidates rejected** for lacking a self-stated method name (R9): "Fusion-Aware Direct 3D Gaussian Generation" (2609.27779) and "Human-AI Collaborative Design Sequence for 3D Assets" (2508.17645) — strict protocol, no exceptions - **CGGS completion**: the pre-existing CGGS entry (ego-centric scene generation, previously `Autonomous Driving`, no arXiv ID) turned out to be the *same paper* as shortlisted 2607.03819 — arXiv ID backfilled, re-categorized into AIGC Content Creation (hence total 934 + 21 new names = **955**, with AIGC holding 22) **Build-side changes**: `scripts/build_knowledge_base.py` — `CANONICAL_CATEGORIES` + "AIGC Content Creation"; `CATEGORY_MAP` + 4 mapping keys (`aigc content creation` / `aigc` / `content creation` / `ai-generated content`). ## New Skill: research-expert-team (17 → 18) `skills/research-expert-team/SKILL.md` — a nine-role full-cycle research expert team: R1 Idea Evaluator · R2 Innovation Miner · R3 Simulated Reviewer · R4 Academic Polisher · R5 AIGC-Trace Hygiene Expert · R6 Framework & Venue Compliance Expert · R7 Format & Typesetting Expert · R8 Experiment Designer · R9 Experiment Analyst - Ships method-selection tables for the **three frontier directions**: Ontology & Knowledge Representation (12), Super Agents & Agentic AI (20), AIGC Content Creation (22) — each anchored to `data/methods.json` as single source of truth (re-query, never from memory) - **AIGC-trace hygiene is scoped to publication hygiene** (source-metadata cleanup, AI-cliché neutralization, style naturalization) with an explicit red line: no evading venue AI-use disclosure policies, no fabricating human-authorship evidence - Orchestrates existing deep assets instead of duplicating: writing templates / claim-verification / Gates / comparison-fairness protocol → `cg-paper-writing`; experiment design → `3dgs-experiment-planner`; comparisons → `3dgs-method-compare`; paper reading → `3dgs-paper-reader`; charts → `3dgs-visualizer` - Dual-copy synced to `~/.workbuddy/skills/` (MD5-verified identical) ## New-Skill Carrier Integration - `README.md` / `README_CN.md`: skills tables 18 rows (row 18 added); What's New / 最新动态 v0.9.10 entry added (v0.9.9 demoted to Previous) - **English category table fully recalibrated**: all 23 listed category counts re-synced to the live build (e.g. Feed-Forward 67→90, Dynamic & 4D 97→93, Embodied AI 30→44, Robustness 11→22, Security 13→11); the two v0.9.9 categories missing from the table (Ontology, Super Agents) added as rows, plus AIGC Content Creation (CN table: 3 rows + 动态与 4D count fix) - `docs/skills.html`: 18th skill card; OpenClaw block + one-click bash script extended (`jaccen-research-expert-team`, "Install all 18 skills") - `docs/methods.html`: OpenClaw block + bash script synced - `.claude-plugin/marketplace.json`: `research-expert-team` mounted in `gaussian-research` (5 + 5 + 8 = 18 across 3 plugins); plugin description extended - `CLAUDE.md`: description Covers-list extended (ontology, super-agent, research expert team) - `references/3dgs-methods-overview.md`: AIGC Content Creation section (22 entries) + Categories Count table → 955 ## Count-Caliber Sync (scripted: `.temp/upgrade_v0910.py`, 131 replacements) - Version 0.9.9 → 0.9.10 (package.json, CLAUDE.md, marketplace.json, app-common.js, index.html) - 934 → 955 / 25 → 26 / 17 → 18 across: README, README_CN, CLAUDE.md, package.json, marketplace metadata, docs (app-common.js i18n zh+en, index/methods/skills/studio HTML meta + badges, ebook descriptive calibers incl. two previously-missed "15 个工程技能" spots), mcp-server README, 7 promo files, 9 SKILL.md descriptions + core-stance.md (678+→955), cg-paper-writing references (826→848 arXiv-holding; 108→107 without IDs), root overview index, academic-submissions list - `scripts/verify_ebook.py`: caliber pattern 934 → 955 - Historical records (changelog/, README update-history sections, references update-log sections, the ebook's 2026.09 timeline) intentionally untouched ## Verification - `scripts/validate_knowledge_base.py`: **PASS** — 955 entries, 848 arXiv-holding (88%), 26-category taxonomy, dedup/fabrication-blacklist checks - `scripts/crosscheck_carriers.py`: **PASS** — json / csv / methods.html / abstracts.js all 955, name sets identical - R9 self-statement check at admission: **22/22 PASS**; post-build category spot-check: AIGC Content Creation = 22, Foundation back to 26 (no fallback leakage), Autonomous Driving 24 (CGGS moved out) ## Not Changed / Deferred - `dist/awesome-gaussian-skills-bundle*` — still the v0.9.8-era snapshot (896/15); rebuild + privacy scan reserved for release packaging - `assets/book-full.png` / `book-fullpage.png` — still show the old 872+ badge (binary screenshots; regenerate when convenient) - Git-history privacy rewrite — deferred by user decision; all future commits clean - **README category tables**: EN table row-level counts now live-synced this round; the CN table lists representative methods per category without counts (by design, unchanged) - New-skill dist packaging — reserved for the next release, per release discipline ## State - All v0.9.10 changes remain **uncommitted** in the working tree — user decision: complete carriers first, commit manually.