# v0.9.8 → v0.9.9 — Knowledge-Base Expansion & Two New Skills (Oct 13, 2026, part 2)

## Summary

Expanded the arXiv-verified knowledge base **900 → 934 methods** and **23 → 25 categories**, shipped **two new skills** (`ontology`, `super-agent`; 15 → 17), remediated a name-collision regression (CGS-SLAM), and unified the site-wide caliber to **934 / 25 / 17** across every carrier. This second same-day entry also records the closing consistency round: the two new skills are now listed in every carrier that enumerates skills.

## Knowledge Base

- **+34 methods admitted** after arXiv candidate search and per-entry verification (`.temp/` script chain: `search_new_categories.py` → `list_candidates.py` → `show_shortlist.py` → `admit_new_methods.py`)
- **2 new categories**: *Ontology & Knowledge Representation* (12 methods) and *Super Agents & Agentic AI* (20 methods); taxonomy 23 → 25
- **CGS-SLAM name-collision remediation**: a newly admitted paper (arXiv:2608.26868) had overwritten the existing CGS-SLAM entry (IROS 2025 / SLAM) in the CSV and `docs/abstracts.js`; the original row and abstract were restored from git HEAD (`fix_cgs.py`, `fix_html_cgs.py`) and the colliding newcomer dropped
- Carrier regeneration: `data/methods.json`, `3dgs-methods-overview.csv`, `docs/methods.html`, `docs/abstracts.js` — **all 934 entries, name sets identical**

## New Skills (15 → 17)

- `skills/ontology/` — ontology & structured knowledge representation for 3D/spatial domains: scene-graph construction (DynamicGSG, GraphWrit3R), ontology-grounded planning (OntoPlan), neuro-symbolic reasoning (SATURN), KG-augmented QA (KLDrive). Backed by the 12-method *Ontology & Knowledge Representation* category.
- `skills/super-agent/` — super-agent & multi-agent LLM system design: architecture patterns (Octopus v2, E-Ledger, SwarmWorld), 3D/spatial agent integration (COLMAR, GRAND-SLAM, CADFather, SPLATIFY), benchmarks & failure diagnosis. Backed by the 20-method *Super Agents & Agentic AI* category.
- Both follow the repo skill contract (frontmatter + KB-backed exemplar tables + red lines + related skills); dual-copy synced to `~/.workbuddy/skills/` (MD5-verified identical).

## New-Skill Carrier Integration (closing round)

- `README.md` / `README_CN.md`: skills tables now list 17 rows (rows 16/17 added); fixed stale `## 15 AI 驱动技能` heading (CN); fixed the book "narrative layer" paragraph (900/15 → 934/17, EN — the CN mirror was already at caliber); added the **v0.9.9 "What's New" / 「最新动态」 entry** (900 → 934, 23 → 25 categories, 15 → 17 skills, links to `changelog/2026-10-13-v2.md`) and demoted the v0.9.8 entry to Previous
- `docs/skills.html`: 2 skill cards added (SKILLS array = 17); meta description 872/15 → 934/17; OpenClaw install block + one-click bash script extended (`jaccen-ontology`, `jaccen-super-agent`; "Install all 17 skills")
- `docs/methods.html`: meta description + OpenClaw block + one-click bash script synced to 17
- `docs/index.html`: hero skill badge 15 → 17 (zh fallback), book-section pills 900 方法/15 项技能 → 934 方法/17 项技能, methods card 23 → 25 categories
- `docs/app-common.js`: `hero.badge.skills` i18n — zh "15 项技能" → "17 项技能", en "15 Skills" → "17 Skills"
- `docs/studio.html`: meta description 872/15 → 934/17
- `package.json`: description "15 skills" → "17 skills"
- `.claude-plugin/marketplace.json`: `gaussian-domain` plugin now mounts `ontology` + `super-agent` (4 + 5 + 8 = 17 across 3 plugins); plugin description extended; stale "900-method" in `gaussian-research` description fixed to 934
- `promo/zhihu-post.md`: "15 个 AI 工程技能" → 17
- `promo/wechat-article-final.md`: 5 stale spots fixed (872 → 934 in 4 places, 16 个 → 17 个 skills, 23 分类 → 25 分类 in the compliance-check data-caliber line)
- `promo/wechat-article.md`: title options + post summary 872/680 → 934
- `promo/wechat-launch.md`: title option 872 → 934
- `.marketplace/awesome-lists/academic-submissions.md`: "900-method 3DGS catalog" → 934

## Count-Caliber Sync (three scripted rounds, same day)

- `upgrade_v099.py` / `_r2` / `_r3`: version 0.9.8 → 0.9.9 everywhere; 900/23/15 → 934/25/17 across README, README_CN, CLAUDE.md, SKILL.md frontmatter/descriptions, docs pages, promo materials, marketplace metadata, and scripts
- Historical records (changelog/, README update-history sections, references update-logs) intentionally untouched

## Verification

- `scripts/validate_knowledge_base.py`: **PASS** — 934 entries, 826 arXiv-holding (88%), taxonomy/dedup/fabrication-blacklist checks
- `scripts/crosscheck_carriers.py`: **PASS** — json / csv / methods.html / abstracts.js all 934, name sets identical
- Stale-count sweep over live files after this round: 0 remaining hits for the 15-skills / 900-method / 23-category calibers (history sections and pre-release `dist/` snapshot excluded)

## Not Changed / Deferred

- `dist/awesome-gaussian-skills-bundle*` — still the v0.9.8-era snapshot (896/15); rebuild + privacy scan (`scripts/finish_skill_release.py`) reserved for the v0.9.9 release packaging, per release discipline
- `assets/book-full.png` / `book-fullpage.png` — still show the old 872+ badge (binary screenshots; regenerate when convenient)
- `skills/cg-paper-writing/references/3dgs-domain-venues.md` — its "数据口径" line is a **data-provenance marker** ("`data/methods.json` (v0.9.6, 881 方法 / 23 类)"): the doc's 23-class distribution tables were computed on that snapshot, so the marker is kept as-is (regenerating the doc against v0.9.9 data is separate content work)
- `skills/cg-paper-writing/references/cjig-format.md` — "1006-8961" is an ISSN, not a count (false positive)
- Git-history privacy rewrite — deferred by user decision; all future commits are clean
- New-skill dist packaging (per-skill zips) — reserved for the next release, per release discipline

## State

- All v0.9.9 changes (this round included) remain **uncommitted** in the working tree — user decision: complete carriers first, commit manually.