Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/verymath/ai4math-optimization/copt-linear-programnpx skills add VeryMath/AI4Math-Optimization --skill copt-linear-programgit clone --depth 1 https://github.com/VeryMath/AI4Math-OptimizationWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00090 | $0.02869 |
| Opus 5 | $0.00045 | $0.01435 |
| Sonnet 5 | $0.00018 | $0.00574 |
| Haiku 4.5 | $0.00009 | $0.00287 |
Grade A, and why
copt-linear-program scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 209 lines — stays where its author put it; the contents beside it link to each section on GitHub.
用 COPT 求解线性规划（Linear Program, LP）
适用场景
- 线性目标：
min或max的c^T x - 线性约束：
A_ub x <= b_ub、A_eq x == b_eq（可只用其中一类或组合） - 变量界：每个
x_j可有下界、上界（或无界）
输入：可以是自然语言/应用题，也可以是已给定的系数矩阵或 JSON。不要默认要求用户先整理成矩阵；仅在用户已提供矩阵或完成符号化后，再用 JSON 字段复现或调用 solve_lp。
Quick Start（先做这个）
按下面清单执行并在回答中保留结构：
- 路径判断：用户给的是自然语言，还是已给矩阵/JSON
- 输出重述（1-2 句）
- 请用户确认重述是否正确（若有问题则澄清，无误则继续）
- 列变量/目标/约束（符号化）
- 需要时提关键澄清问题，或明确写出假设
- 给出求解结果（目标值 + 变量值）
- 用 1-2 句解释业务含义
执行流程（两条路径）
flowchart LR
A[Matrix_or_JSON]
B[Natural_language]
A --> A1[Build_or_solve_lp]
B --> B1[Restate]
B1 --> B2[User_confirm]
B2 --> B3[Clarify_or_assume]
B3 --> B4[Symbolic_model]
B4 --> B5[Numeric_or_addVar]
B5 --> B6[Solve_and_report]
A1 --> B6
路径 A：已有矩阵或 JSON
- 核对维度：
c长度、A行列、b长度与约束条数一致。 - 用 scripts/solve_lp.py 中的
solve_lp，或按「手建模型」自行写coptpy（适合需要命名变量、稀疏结构的题）。
路径 B：自然语言 / 应用题
用户未给数字矩阵时，Agent 不要先索要 JSON。按下面交付物顺序推进：
| 步骤 | 内容 |
|---|---|
| 1. 重述 | 用一两句话复述题意，便于用户确认理解是否正确。 |
| 2. 用户确认 | 请用户确认重述是否准确，如有偏差则澄清；无异议再继续。 |
| 3. 符号化 | 变量表：名称、含义、单位（若有）、是否非负。目标：min 还是 max，线性式。约束：逐条写出，并标明 <= / >= / =（注意「不超过」「不少于」与不等号方向）。 |
| 4. 数值化 | 把符号模型写成 c、A_ub/b_ub、A_eq/b_eq、bounds；或跳过稠密矩阵，用 addVar + addConstr + 有意义约束名直接建模。 |
| 5. 求解与回答 | 给出最优值、各变量取值；必要时用一句话解释经济/物理含义（用户未问不必展开对偶或灵敏度）。 |
输出模板（推荐）
回答尽量按以下模板组织（可省略不适用小节）：
### 问题重述
...
### 符号化模型
- 决策变量：...
- 目标函数：...
- 约束：...
### 数值化（可选）
- c: ...
- A_ub / b_ub: ...
- A_eq / b_eq: ...
- bounds: ...
### 求解结果
- status: ...
- objective: ...
- x: ...
### 结果解释
...
歧义与澄清
建模前若信息不足，优先提问；若是标准教材题型，可列出假设再求解，并在重述中写明假设：
- 目标是最小化成本/资源还是最大化利润/效用（口语可能含糊）。
- 「至多 / 不超过」→ 通常对应
<=；「至少 / 不少于」→ 通常对应>=（按变量所在一侧核对）。 - 是否允许分数解（连续 LP 默认可行）；若题意明确要求整数台数、0-1 选址等，见下节，勿当连续变量悄悄求解。
- 非负：产量、投入量等若题中未写，常默认
>=0，须在变量表里写明「假设非负」。 - 多商品、多期、多维约束：检查下标与矩阵行、列是否一一对应，避免把行维与列维弄反。
范围与非 LP（本 skill 边界）
本文件针对连续变量的 LP。若叙述中出现下列情况，应向用户说明已超出纯 LP，需 MILP、非线性或其它建模，禁止不声明就把整数松弛成连续变量：
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 209 lines · 90 tokens per session scan A 1337f4ee7ad4
copt-linear-program is a skill published in the GitHub repository VeryMath/AI4Math-Optimization (5 stars, last pushed 1mo ago), licensed MIT. It adds 90 tokens to every session and 2,869 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
math-beamer
Use when creating, revising, or auditing source-grounded mathematical Beamer slide decks from papers, notes, proofs, lectures, experiments, or existing LaTeX slides across pure mathematics, applied mathematics, computational mathematics, statistics, optimization, geometry, topology, algebra, number theory…
paper-writing
Use when drafting, revising, or auditing source-grounded mathematical paper text, including abstracts, introductions, related work, theorem exposition, experiment narratives, revision plans, and response letters.
latex-build-and-layout-audit
Use when checking LaTeX paper projects for compilation, latexmk logs, undefined references, citation issues, duplicate labels, macro/package hygiene, layout warnings, floats, arXiv, or venue compatibility.
proof-obligation-and-assumption-audit
Use when checking mathematical paper results for assumptions, quantifiers, domains, dependency fit, edge cases, external theorem use, proof coverage, or theorem-to-claim consistency.
claim-evidence-ledger
Use when auditing mathematical paper drafts for supported claims, missing citations, overclaims, proof status, experiment support, or source-to-text traceability.
paper-skeleton-and-logical-architecture
Use when turning mathematical notes, theorem statements, proof sketches, experiments, or reading outputs into a paper skeleton, section plan, result dependency map, or contribution architecture before prose drafting.