VeryMath · ConanXu-math · Sep 2, 2026 · Sep 2, 2026 · Sep 2, 2026 · Sep 2, 2026
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
@@ -17,6 +17,7 @@ attribution ledger.
 | --- | --- | --- | --- | --- |
 | Root repository router | Current `VeryMath/AI4Math-Optimization` repository | [Conan Xu](https://github.com/ConanXu-math), conanxu <1845830029@qq.com> | Root README, bilingual repository map, package routing layer, and public repository structure. | Preserved through repository commit history. |
 | `skills/osqp-solver` | New package prepared in the AI4Math Skill Library from official OSQP documentation | [Conan Xu](https://github.com/ConanXu-math) | Package commissioning, integration, review, and release stewardship; implementation was AI-assisted and independently tested. | Preserve through the upstream integration commit and pull request. |
+| `skills/optskills` | [fujiwaranoM0kou/OptSkills](https://github.com/fujiwaranoM0kou/OptSkills) | [Haochen Yang](https://github.com/fujiwaranoM0kou) and the OptSkills paper authors | Released optimization problem-archetype skill cards; VeryMath adds standalone packaging, indexing, documentation, and update instructions. | Original MIT notice is preserved in `skills/optskills/LICENSE`; source scope is recorded in `skills/optskills/SOURCES.md`; import authorship is preserved in the import commit. |
 | Optimization skill packages | Repository history and imported optimization work | [ricercar77](https://github.com/ricercar77) | Optimization skill contributions and package content. | Preserved through GitHub contributor graph, repository history, and this ledger. |
 | `skills/linear-programming`, `skills/mixed-integer-programming`, `skills/second-order-cone-programming`, `skills/or-solver` | Current package metadata | 李爽夕 | LP, MIP, SOCP, solver-setup workflow content, examples, and benchmark materials. | Recorded in package metadata and this ledger. |
 

diff --git a/README.md b/README.md
@@ -9,7 +9,7 @@ manifold-constrained optimization.
 [中文说明](README.zh-CN.md) · [Contributors](CONTRIBUTORS.md) · [Skill packages](#skill-packages) · [Installation](#installation) · [Quick start](#quick-start) · [Security model](#security-and-scope)
 
 ![version](https://img.shields.io/badge/version-0.1.0-blue)
-![skills](https://img.shields.io/badge/skills-7-2ea44f)
+![skills](https://img.shields.io/badge/skills-8-2ea44f)
 ![license](https://img.shields.io/badge/license-MIT-green)
 
 </div>
@@ -39,6 +39,9 @@ package that matches the problem class.
 | [`second-order-cone-programming`](skills/second-order-cone-programming/) | SOCP modeling and cvxpy-based conic solver workflows. | [`README`](skills/second-order-cone-programming/README.md) · [`SKILL`](skills/second-order-cone-programming/SKILL.md) |
 | [`or-solver`](skills/or-solver/) | Shared solver detection, installation planning, license checks, and solver selection for OR skills. | [`README`](skills/or-solver/README.md) · [`SKILL`](skills/or-solver/SKILL.md) |
 | [`osqp-solver`](skills/osqp-solver/) | OSQP modeling, repeated solves, status gates, and independent verification for continuous convex QPs. | [`README`](skills/osqp-solver/README.md) · [`SKILL`](skills/osqp-solver/SKILL.md) |
+| [`optskills`](skills/optskills/) | Standalone selection and use of 103 released OptSkills problem-archetype cards, with explicit upstream updates. | [`README`](skills/optskills/README.md) · [`SKILL`](skills/optskills/SKILL.md) |
+
+`optskills` is self-contained and does not require sibling packages.
 
 ## Installation
 
@@ -57,6 +60,7 @@ Skill paths:
 - skills/second-order-cone-programming
 - skills/or-solver
 - skills/osqp-solver
+- skills/optskills
 
 Steps:
 1. Clone or update the repository locally.
@@ -80,6 +84,7 @@ ln -s "$PWD/skills/mixed-integer-programming" ~/.codex/skills/mixed-integer-prog
 ln -s "$PWD/skills/second-order-cone-programming" ~/.codex/skills/second-order-cone-programming
 ln -s "$PWD/skills/or-solver" ~/.codex/skills/or-solver
 ln -s "$PWD/skills/osqp-solver" ~/.codex/skills/osqp-solver
+ln -s "$PWD/skills/optskills" ~/.codex/skills/optskills
 ```
 
 If your agent uses a different local Skill directory, replace `~/.codex/skills` with that configured path.
@@ -112,6 +117,12 @@ For continuous convex QPs solved with OSQP, start with:
 skills/osqp-solver/SKILL.md
 ```
 
+For natural-language operations-research archetypes, start with:
+
+```text
+skills/optskills/SKILL.md
+```
+
 ## Repository Layout
 
 ```text
@@ -126,6 +137,7 @@ AI4Math-Optimization/
     ├── mixed-integer-programming/
     ├── or-solver/
     ├── osqp-solver/
+    ├── optskills/
     └── second-order-cone-programming/
 ```
 

diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -8,7 +8,7 @@
 [English](README.md) · [贡献者](CONTRIBUTORS.md) · [技能包](#技能包) · [安装](#安装) · [快速开始](#快速开始) · [安全边界](#安全边界)
 
 ![version](https://img.shields.io/badge/version-0.1.0-blue)
-![skills](https://img.shields.io/badge/skills-7-2ea44f)
+![skills](https://img.shields.io/badge/skills-8-2ea44f)
 ![license](https://img.shields.io/badge/license-MIT-green)
 
 </div>
@@ -35,6 +35,9 @@
 | [`second-order-cone-programming`](skills/second-order-cone-programming/) | SOCP 建模和基于 cvxpy 的锥优化求解流程。 | [`README`](skills/second-order-cone-programming/README.md) · [`SKILL`](skills/second-order-cone-programming/SKILL.md) |
 | [`or-solver`](skills/or-solver/) | 为 OR skills 提供统一求解器检测、安装规划、license 检查和选择策略。 | [`README`](skills/or-solver/README.md) · [`SKILL`](skills/or-solver/SKILL.md) |
 | [`osqp-solver`](skills/osqp-solver/) | 面向连续凸 QP 的 OSQP 建模、重复求解、状态门槛与独立验证。 | [`README`](skills/osqp-solver/README.zh-CN.md) · [`SKILL`](skills/osqp-solver/SKILL.md) |
+| [`optskills`](skills/optskills/) | 独立选择和使用 OptSkills 已发布的 103 个优化问题原型知识卡，并支持显式的上游更新。 | [`README`](skills/optskills/README.zh-CN.md) · [`SKILL`](skills/optskills/SKILL.md) |
+
+`optskills` 是自包含的，不依赖相邻技能包。
 
 ## 安装
 
@@ -53,6 +56,7 @@ Skill 路径：
 - skills/second-order-cone-programming
 - skills/or-solver
 - skills/osqp-solver
+- skills/optskills
 
 请执行：
 1. 本地 clone 或更新仓库。
@@ -76,6 +80,7 @@ ln -s "$PWD/skills/mixed-integer-programming" ~/.codex/skills/mixed-integer-prog
 ln -s "$PWD/skills/second-order-cone-programming" ~/.codex/skills/second-order-cone-programming
 ln -s "$PWD/skills/or-solver" ~/.codex/skills/or-solver
 ln -s "$PWD/skills/osqp-solver" ~/.codex/skills/osqp-solver
+ln -s "$PWD/skills/optskills" ~/.codex/skills/optskills
 ```
 
 如果你的 agent 使用别的本地 Skill 目录，把 `~/.codex/skills` 替换成对应配置路径。
@@ -107,6 +112,12 @@ skills/cdopt-optimization/SKILL.md
 skills/osqp-solver/SKILL.md
 ```
 
+自然语言运筹优化问题原型从这里开始：
+
+```text
+skills/optskills/SKILL.md
+```
+
 ## 仓库结构
 
 ```text
@@ -121,6 +132,7 @@ AI4Math-Optimization/
     ├── mixed-integer-programming/
     ├── or-solver/
     ├── osqp-solver/
+    ├── optskills/
     └── second-order-cone-programming/
 ```
 

diff --git a/SKILL.md b/SKILL.md
@@ -20,6 +20,10 @@ Use this repository as a routing layer for optimization workflows.
 - `skills/or-solver/`: shared optimization solver setup and selection.
 - `skills/osqp-solver/`: continuous convex QP modeling, OSQP execution,
   status gates, and independent solution or infeasibility-certificate checks.
+- `skills/optskills/`: standalone selection and use of released OptSkills
+  problem-archetype cards. Read its `SKILL.md`, combined
+  `skill_library/index.json`, and selected card files. It does not depend on
+  sibling skill packages.
 
 Prefer package-local instructions over this router when running a concrete
 workflow.
diff --git a/skills/optskills/LICENSE b/skills/optskills/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Haochen Yang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/skills/optskills/README.md b/skills/optskills/README.md
@@ -0,0 +1,50 @@
+# OptSkills
+
+OptSkills is a standalone thin entrypoint to 103 released OptSkills
+problem-archetype cards. It helps an agent select a card, formulate the
+user's actual operations-research problem, solve it when the environment
+permits, and report what was checked.
+
+## Source selection
+
+The library contains 93 NanoCO cards and 10 learned-only cards. It includes no
+cluster duplicate and no `ingredients.json`. The selected cards and their
+relative paths are listed in `skill_library/index.json`.
+
+## Installation
+
+Clone `VeryMath/AI4Math-Optimization`, then link only `skills/optskills` into
+the agent's skill directory:
+
+```bash
+git clone https://github.com/VeryMath/AI4Math-Optimization.git
+cd AI4Math-Optimization
+mkdir -p ~/.codex/skills
+ln -s "$PWD/skills/optskills" ~/.codex/skills/optskills
+```
+
+## Quick start
+
+Use this prompt:
+
+> Read `SKILL.md`, select the relevant OptSkills cards through
+> `skill_library/index.json`, formulate my problem, solve it if possible, and
+> check the key constraints.
+
+## Updates
+
+Use this prompt only when an upstream update is wanted:
+
+> Update this standalone OptSkills package from the official upstream. Read `UPDATE.md`, report proposed changes first, and wait for approval before editing.
+
+## Boundaries
+
+Raw cards may contain placeholders or unverified examples. Being installed,
+loaded, solved, and checked are different claims; report each state separately.
+Normal use is package-local and does not call sibling skills or require the
+upstream training, agent, chat, or embedding system.
+
+## License and sources
+
+See [SOURCES.md](SOURCES.md) for upstream attribution and [LICENSE](LICENSE)
+for licensing information.
diff --git a/skills/optskills/README.zh-CN.md b/skills/optskills/README.zh-CN.md
@@ -0,0 +1,43 @@
+# OptSkills
+
+OptSkills 是一个独立的薄入口，连接 103 张已发布的 OptSkills
+问题原型卡片。它帮助智能体选择卡片、为用户的实际运筹问题建模、在环境允许时求解，并如实报告已检查的内容。
+
+## 来源选择
+
+该库包含 93 张 NanoCO 卡片和 10 张仅来自 learned 的卡片；不包含重复的
+cluster 卡片，也不包含 `ingredients.json`。选中的卡片及其相对路径列在
+`skill_library/index.json` 中。
+
+## 安装
+
+克隆 `VeryMath/AI4Math-Optimization`，然后只把 `skills/optskills`
+链接到智能体的技能目录：
+
+```bash
+git clone https://github.com/VeryMath/AI4Math-Optimization.git
+cd AI4Math-Optimization
+mkdir -p ~/.codex/skills
+ln -s "$PWD/skills/optskills" ~/.codex/skills/optskills
+```
+
+## 快速开始
+
+使用下面的提示：
+
+> 阅读 `SKILL.md`，通过 `skill_library/index.json` 选择相关的 OptSkills
+> 卡片，为我的问题建模，在可以时求解，并检查关键约束。
+
+## 更新
+
+只有需要上游更新时才使用下面的提示：
+
+> Update this standalone OptSkills package from the official upstream. Read `UPDATE.md`, report proposed changes first, and wait for approval before editing.
+
+## 边界
+
+原始卡片可能含有占位符或未经验证的示例。已安装、已加载、已求解和已检查是不同的声明；应分别报告。普通使用仅依赖本包，不调用同仓库的其他技能，也不要求上游的训练、智能体、聊天或嵌入系统。
+
+## 许可与来源
+
+上游署名见 [SOURCES.md](SOURCES.md)，许可信息见 [LICENSE](LICENSE)。
diff --git a/skills/optskills/SKILL.md b/skills/optskills/SKILL.md
@@ -0,0 +1,53 @@
+---
+name: optskills
+description: Use when a coding agent must model or solve a natural-language operations-research problem by selecting from the released OptSkills problem-archetype cards, or when the user explicitly asks to update this standalone OptSkills library from its official upstream repository.
+---
+
+# OptSkills
+
+Use this package independently. Do not call sibling skills in the enclosing
+repository and do not require the OptSkills training, agent, chat, or embedding
+system.
+
+## Lightweight version check
+
+At the start of every normal use, read only the `Upstream snapshot commit` line
+from `SOURCES.md` and run
+`git ls-remote https://github.com/fujiwaranoM0kou/OptSkills.git HEAD`. Compare
+only those two commits; do not fetch or inspect upstream cards. If they match,
+continue silently. If they differ or the lookup fails, give at most one short
+note and continue with the packaged cards without retrying. Read `UPDATE.md`
+and inspect upstream content only when the user asks to update.
+
+## Normal use
+
+1. Restate the user's sets, parameters, decision variables, objective,
+   constraints, units, and assumptions.
+2. Read `skill_library/index.json`. Compare the problem structure with each
+   entry's `name` and `description`, then select one to three candidates.
+3. Read only the selected files using their relative `path` fields. Never
+   invent a path from a skill ID.
+4. Choose the closest card by variables, objective, and constraint structure.
+   If none fits, say that no released card matched and continue with general
+   mathematical modeling without claiming a card was used.
+5. Apply the card to the user's actual data. Card examples are guidance, not
+   proof that the new model is correct.
+6. Use only solver libraries and licenses actually available in the current
+   environment. Ask before installing dependencies.
+7. Check solver status before reading values. Recompute the objective and the
+   problem's key constraints whenever execution is possible.
+8. Report separately: selected card, completed model, solver execution,
+   constraint checks, and unresolved limitations.
+
+## Explicit update requests
+
+Only when the user asks to update or synchronize OptSkills, read `UPDATE.md`
+and follow it. Do not update during an ordinary modeling task.
+
+## Source boundary
+
+The files under `skill_library/` are released upstream material. Do not rewrite
+them as part of normal use. If a card contains a placeholder, missing
+dependency, inconsistent notation, or suspected error, correct the active task
+explicitly, report the issue, and leave the card unchanged until an upstream
+release replaces it.
diff --git a/skills/optskills/SOURCES.md b/skills/optskills/SOURCES.md
@@ -0,0 +1,31 @@
+# Sources and attribution
+
+This package provides a thin, standalone entrypoint to released optimization
+skill cards from [OptSkills](https://github.com/fujiwaranoM0kou/OptSkills).
+
+Upstream snapshot commit: d9e14300df4b499529c74ea1981e2c1aba0628b8
+
+Paper: *OptSkills: Learning Generalizable Optimization Skills from Problem
+Archetypes via Cluster-Based Distillation*, arXiv:2605.29829.
+
+Paper authors: Haochen Yang, Ke Zhao, Mengyuan Ma, Xingyu Lu, Xiangfeng Wang,
+and Hong Qian.
+
+## Included material
+
+- all 93 Markdown cards from `skill_library_nanoco_learned`;
+- the 10 `skill_library_learned` cards whose `skill_id` is absent from NanoCO;
+- the selected entries from the two upstream `index.json` files.
+
+The Markdown card bodies and selected index fields are kept as published
+upstream. VeryMath adds only the standalone entrypoint, combined index, human
+documentation, and reviewed update instructions.
+
+## Excluded material
+
+The OptSkills training, clustering, self-learning, trajectory, evaluation,
+agent, LLM, embedding, dataset, and `ingredients.json` components are not part
+of this package and are not runtime dependencies.
+
+The included upstream material is distributed under the MIT license in
+`LICENSE`. Consult the source repository and paper for the original project.
diff --git a/skills/optskills/UPDATE.md b/skills/optskills/UPDATE.md
@@ -0,0 +1,53 @@
+# Updating the standalone OptSkills library
+
+Run this workflow only after the user explicitly asks for an upstream update.
+
+1. Read `SOURCES.md`, this file, and the current `skill_library/index.json`.
+2. Fetch the official `fujiwaranoM0kou/OptSkills` repository.
+3. Inspect the current NanoCO, learned, and cluster indexes. If their meaning,
+   fields, or directory structure changed, stop and explain the change before
+   applying the old selection rule.
+4. Before any edit, preflight each selected upstream index and every file it
+   names. Within each index, require unique non-empty `skill_id` values and
+   unique `path` values. Accept a path only when it is a direct-child Markdown
+   filename under that source library. Reject absolute paths, separators or
+   nested paths, `.` or `..`, non-`.md` suffixes, missing files, and symlinks.
+   Never follow an upstream path outside its source library.
+5. Form the new selected set in this order: NanoCO first, learned only for IDs
+   absent from NanoCO, and cluster only for IDs absent from both when cluster is
+   still published as directly usable cards. Require the selected set itself
+   to have unique non-empty IDs and unique paths.
+6. Complete the comparison before writing. Compare the selected set with the
+   local package by `skill_id` and ordinary text review. Report every addition,
+   body change, source switch, path change, full upstream removal, path
+   collision, and license or attribution change. Never rely only on the index
+   `version` field.
+7. Treat an ID moving from NanoCO to learned or cluster as a source switch, not
+   a removal. Ask again for every current ID removed from all three upstream
+   libraries. Resolve every approval before editing; if any removal is
+   declined, leave the entire package unchanged.
+8. Compare every proposed path with all currently occupied local paths,
+   including paths owned by IDs that will move or disappear. A path occupied
+   by a different current ID remains a collision until that ID's move or
+   removal is approved. Stop on any unresolved collision, and never overwrite
+   a current card while its removal is unresolved.
+9. Treat any upstream license, copyright-holder, or attribution change as a
+   separate stop. Do not import that release until redistribution compatibility
+   is confirmed and the user explicitly approves it. If approved, update
+   `LICENSE`, `SOURCES.md`, root `CONTRIBUTORS.md`, and any affected README
+   claims together.
+10. After all comparisons and approvals are complete, copy selected card text
+    and index fields as published upstream. Follow each accepted upstream path;
+    when an ID moves, remove its obsolete path. Update every affected card plus
+    index, source, count wording, and the `SOURCES.md` upstream snapshot commit
+    in `index.json`, `SOURCES.md`, and any affected package or root README files
+    in the same reviewable Git change.
+11. Validate every index path and run real problems for every affected card.
+    Report file updates separately from solver execution and checked results.
+    If final validation fails, do not publish the change; restore the pre-update
+    state through the repository's normal review/rollback workflow. Do not
+    create an old-content backup directory.
+
+Do not add a synchronization script, retain old card copies, rewrite card
+content into a local style, modify sibling skill packages, or add integrity,
+receipt, manifest, authorization, or execution-unlock mechanisms.