VeryMath · ConanXu-math · Jul 28, 2026 · Jul 28, 2026
diff --git a/.github/workflows/validate.yml b/.github/workflows/validate.yml
@@ -0,0 +1,31 @@
+name: Validate skills
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: read
+
+jobs:
+  unittest:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version:
+          - "3.10"
+          - "3.13"
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+          cache-dependency-path: skills/osqp-solver/requirements-test.txt
+      - name: Install OSQP test dependency
+        run: python -m pip install -r skills/osqp-solver/requirements-test.txt
+      - name: Run repository tests with zero skips
+        run: python -m unittest discover -s tests -v
diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
@@ -16,6 +16,7 @@ attribution ledger.
 | Module | Source | Contributor(s) | Contribution scope | Attribution record |
 | --- | --- | --- | --- | --- |
 | 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. |
 | 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
@@ -2,13 +2,14 @@
 
 # AI4Math · Optimization
 
-Skill packages for mathematical optimization modeling, solver setup, LP/MIP/SOCP
-workflows, and manifold-constrained optimization.
+Skill packages for mathematical optimization modeling, solver setup,
+LP/MIP/SOCP workflows, continuous convex QP/OSQP, and
+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-6-2ea44f)
+![skills](https://img.shields.io/badge/skills-7-2ea44f)
 ![license](https://img.shields.io/badge/license-MIT-green)
 
 </div>
@@ -37,6 +38,7 @@ package that matches the problem class.
 | [`mixed-integer-programming`](skills/mixed-integer-programming/) | MILP/MIP modeling with binary, integer, and continuous decision variables. | [`README`](skills/mixed-integer-programming/README.md) · [`SKILL`](skills/mixed-integer-programming/SKILL.md) |
 | [`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) |
 
 ## Installation
 
@@ -54,6 +56,7 @@ Skill paths:
 - skills/mixed-integer-programming
 - skills/second-order-cone-programming
 - skills/or-solver
+- skills/osqp-solver
 
 Steps:
 1. Clone or update the repository locally.
@@ -76,6 +79,7 @@ ln -s "$PWD/skills/linear-programming" ~/.codex/skills/linear-programming
 ln -s "$PWD/skills/mixed-integer-programming" ~/.codex/skills/mixed-integer-programming
 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
 ```
 
 If your agent uses a different local Skill directory, replace `~/.codex/skills` with that configured path.
@@ -102,6 +106,12 @@ For general LP/MIP/SOCP modeling, start with the matching package under
 skills/cdopt-optimization/SKILL.md
 ```
 
+For continuous convex QPs solved with OSQP, start with:
+
+```text
+skills/osqp-solver/SKILL.md
+```
+
 ## Repository Layout
 
 ```text
@@ -115,6 +125,7 @@ AI4Math-Optimization/
     ├── linear-programming/
     ├── mixed-integer-programming/
     ├── or-solver/
+    ├── osqp-solver/
     └── second-order-cone-programming/
 ```
 

diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -2,12 +2,13 @@
 
 # AI4Math · 优化
 
-面向数学优化建模、求解器配置、LP/MIP/SOCP workflow 和流形约束优化的 AI4Math 技能集合。
+面向数学优化建模、求解器配置、LP/MIP/SOCP workflow、连续凸 QP/OSQP
+和流形约束优化的 AI4Math 技能集合。
 
 [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-6-2ea44f)
+![skills](https://img.shields.io/badge/skills-7-2ea44f)
 ![license](https://img.shields.io/badge/license-MIT-green)
 
 </div>
@@ -33,6 +34,7 @@
 | [`mixed-integer-programming`](skills/mixed-integer-programming/) | 含 binary、integer 和 continuous 变量的 MILP/MIP 建模。 | [`README`](skills/mixed-integer-programming/README.md) · [`SKILL`](skills/mixed-integer-programming/SKILL.md) |
 | [`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) |
 
 ## 安装
 
@@ -50,6 +52,7 @@ Skill 路径：
 - skills/mixed-integer-programming
 - skills/second-order-cone-programming
 - skills/or-solver
+- skills/osqp-solver
 
 请执行：
 1. 本地 clone 或更新仓库。
@@ -72,6 +75,7 @@ ln -s "$PWD/skills/linear-programming" ~/.codex/skills/linear-programming
 ln -s "$PWD/skills/mixed-integer-programming" ~/.codex/skills/mixed-integer-programming
 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
 ```
 
 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。
@@ -97,6 +101,12 @@ skills/or-solver/SKILL.md
 skills/cdopt-optimization/SKILL.md
 ```
 
+连续凸 QP 与 OSQP workflow 从这里开始：
+
+```text
+skills/osqp-solver/SKILL.md
+```
+
 ## 仓库结构
 
 ```text
@@ -110,6 +120,7 @@ AI4Math-Optimization/
     ├── linear-programming/
     ├── mixed-integer-programming/
     ├── or-solver/
+    ├── osqp-solver/
     └── second-order-cone-programming/
 ```
 

diff --git a/SKILL.md b/SKILL.md
@@ -18,6 +18,8 @@ Use this repository as a routing layer for optimization workflows.
 - `skills/mixed-integer-programming/`: MIP and MILP modeling workflows.
 - `skills/second-order-cone-programming/`: SOCP modeling and solver 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.
 
 Prefer package-local instructions over this router when running a concrete
 workflow.
diff --git a/pyproject.toml b/pyproject.toml
@@ -11,6 +11,7 @@ dependencies = []
 
 [project.optional-dependencies]
 dev = ["pytest>=8.0"]
+osqp = ["osqp>=1,<2"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests"]
diff --git a/skills/osqp-solver/README.md b/skills/osqp-solver/README.md
@@ -0,0 +1,115 @@
+# OSQP Solver Skill
+
+Chinese guide: [README.zh-CN.md](README.zh-CN.md)
+
+This AI4Math skill models, solves, updates, and independently checks continuous
+convex quadratic programs with the native OSQP Python interface. It is a
+release-ready local candidate for
+[`VeryMath/AI4Math-Optimization`](https://github.com/VeryMath/AI4Math-Optimization);
+it is not an official upstream package until that repository accepts and merges
+it.
+
+## Scope
+
+Use the skill for
+
+\[
+\min_x \tfrac12 x^\top P x + q^\top x
+\quad\text{subject to}\quad l \leq Ax \leq u,
+\qquad P=P^\top\succeq0.
+\]
+
+Do not route mixed-integer, nonconvex, nonlinear-constraint, SDP, or general
+SOCP models to this package. The bundled JSON runner is an auditable tool for
+small dense examples, not a production sparse-data format.
+
+## Isolated installation
+
+Create an isolated environment and install the supported OSQP major version:
+
+```bash
+python3 -m venv .venv-osqp
+source .venv-osqp/bin/activate
+python -m pip install -r requirements-test.txt
+python scripts/check_osqp_environment.py --json
+```
+
+The probe reports `ready: true` only when `osqp`, `numpy`, and `scipy` import
+successfully and the OSQP version is in the supported range. Obtain approval
+before installing into an existing environment.
+
+## Quick start
+
+From this package directory:
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json
+python scripts/solve_qp.py references/primal-infeasible-example.json
+python scripts/solve_qp.py references/dual-infeasible-example.json
+```
+
+Use `--validate-only` for dependency-free structural validation. It does not
+run PSD or numerical solution checks:
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json --validate-only
+```
+
+For a persistent report, choose a path different from the input:
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json \
+  --output outputs/feasible-result.json
+```
+
+## CLI contract and limits
+
+- Successful validation or an accepted solve exits `0`. Input/dependency
+  failures exit `2`; a rejected numerical status or solver failure exits `3`.
+- Standard output is exactly one JSON document on success. Machine-readable
+  errors are written as exactly one JSON document to standard error.
+- Result documents use `schema_version: osqp-solver-result-v1`; structural-only
+  reports use `osqp-solver-validation-v1`.
+- `settings.verbose: true` is rejected because OSQP logging would corrupt the
+  single-JSON standard-output contract.
+- The runner rejects resolved input/output paths that are the same. A distinct
+  existing output file may be atomically replaced, so use a run-specific path.
+- The default dense guard accepts at most a 32 MiB input, `n <= 1000`, and
+  `2,000,000` logical entries across `P` and `A`. Use the native sparse API for
+  larger problems.
+- An accepted solution must pass primal feasibility, stationarity,
+  normal-cone/complementarity, objective, and PSD checks. Infeasibility
+  certificates are normalized and must pass scale-aware residual, sign, and
+  strict negative-margin checks.
+
+Read [SKILL.md](SKILL.md) for the workflow and
+[references/verification.md](references/verification.md) for the mathematical
+acceptance conditions.
+
+## Validation
+
+From the AI4Math Skill Library root, with the isolated environment active:
+
+```bash
+python -m unittest discover -s tests -p 'test_osqp_solver_skill.py' -v
+python scripts/validate_skill_repo.py
+```
+
+Do not treat skipped numerical tests as release evidence. The OSQP-enabled
+suite must run without skips before publishing; inspect the unittest summary
+and require `skipped=0`.
+
+## License, citation, and provenance
+
+This package is intended to follow the AI4Math repository's MIT license after
+maintainer authorization and upstream acceptance. The separately installed
+OSQP project is Apache-2.0 licensed; this package does not vendor or relicense
+OSQP code. Software licensing and academic citation are separate obligations.
+Use [references/citation.md](references/citation.md) for the official OSQP
+citation.
+
+The workflow and references were prepared from the official OSQP documentation
+linked in [references/python-api.md](references/python-api.md). Before public
+release, a human maintainer must confirm publication authorization and record
+factual contributor attribution. Packaging by an AI agent is not that
+authorization.
diff --git a/skills/osqp-solver/README.zh-CN.md b/skills/osqp-solver/README.zh-CN.md
@@ -0,0 +1,101 @@
+# OSQP Solver Skill
+
+English guide: [README.md](README.md)
+
+这个 AI4Math skill 使用原生 OSQP Python 接口，对连续凸二次规划进行建模、求解、
+参数更新和独立复核。它目前是
+[`VeryMath/AI4Math-Optimization`](https://github.com/VeryMath/AI4Math-Optimization)
+的发布就绪本地候选；只有目标仓库接受并合并后，才能称为官方上游 package。
+
+## 适用范围
+
+使用本 skill 处理
+
+\[
+\min_x \tfrac12 x^\top P x + q^\top x
+\quad\text{subject to}\quad l \leq Ax \leq u,
+\qquad P=P^\top\succeq0.
+\]
+
+不要把混合整数、非凸、非线性约束、SDP 或一般 SOCP 模型路由到本 package。
+附带的 JSON runner 用于可审查的小型 dense 示例，不是生产级 sparse 数据格式。
+
+## 隔离安装
+
+创建隔离环境并安装受支持的 OSQP 主版本：
+
+```bash
+python3 -m venv .venv-osqp
+source .venv-osqp/bin/activate
+python -m pip install -r requirements-test.txt
+python scripts/check_osqp_environment.py --json
+```
+
+只有 `osqp`、`numpy`、`scipy` 均可导入，而且 OSQP 版本落在支持范围内时，探针才会
+报告 `ready: true`。修改已有环境前先取得批准。
+
+## 快速开始
+
+在本 package 目录运行：
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json
+python scripts/solve_qp.py references/primal-infeasible-example.json
+python scripts/solve_qp.py references/dual-infeasible-example.json
+```
+
+`--validate-only` 可在不安装数值依赖时做结构校验，但不会执行 PSD 或数值解检查：
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json --validate-only
+```
+
+需要保存报告时，输出路径必须与输入路径不同：
+
+```bash
+python scripts/solve_qp.py references/feasible-example.json \
+  --output outputs/feasible-result.json
+```
+
+## CLI 合同与边界
+
+- 结构校验成功或求解结果被接受时退出码为 `0`；输入/依赖错误为 `2`；数值状态未被
+  接受或 solver 失败为 `3`。
+- 成功时标准输出严格为一个 JSON 文档；机器可读错误严格以一个 JSON 文档写入标准错误。
+- 求解报告使用 `schema_version: osqp-solver-result-v1`；仅结构校验的报告使用
+  `osqp-solver-validation-v1`。
+- 拒绝 `settings.verbose: true`，避免 OSQP 日志破坏标准输出的单 JSON 合同。
+- runner 会拒绝解析后相同的输入/输出路径，从而保留输入。不同路径上的已有输出文件
+  可能被原子替换，因此应使用每次运行独立的路径。
+- 默认 dense guard 接受的输入最大为 32 MiB，要求 `n <= 1000`，并且 `P` 与 `A`
+  的 logical entry 总数不超过 `2,000,000`。更大问题应直接使用原生 sparse API。
+- 被接受的解必须通过 primal feasibility、stationarity、
+  normal-cone/complementarity、objective 和 PSD 检查；不可行性证书会先归一化，
+  再接受尺度化 residual、方向符号和严格负裕量检查。
+
+完整工作流见 [SKILL.md](SKILL.md)，数学验收条件见
+[references/verification.md](references/verification.md)。
+
+## 验证
+
+激活隔离环境后，在 AI4Math Skill Library 根目录运行：
+
+```bash
+python -m unittest discover -s tests -p 'test_osqp_solver_skill.py' -v
+python scripts/validate_skill_repo.py
+```
+
+被跳过的数值测试不能作为发布证据；公开发布前必须在安装 OSQP 的环境中做到该测试集
+零跳过，并检查 unittest 汇总中 `skipped=0`。
+
+## 许可证、引用与来源边界
+
+在维护者授权并被上游接受后，本 package 预期沿用 AI4Math 仓库的 MIT license。
+单独安装的 OSQP 项目采用 Apache-2.0 license；本 package 不打包、也不重新许可
+OSQP 代码。软件许可和学术引用是两项不同义务。官方 OSQP 引用见
+[references/citation.md](references/citation.md)。
+
+本工作流和参考材料依据
+[references/python-api.md](references/python-api.md) 所列 OSQP 官方文档整理。
+公开发布前，人类维护者必须确认发布授权，并如实记录贡献者署名；AI agent 完成打包
+不能替代这项授权。