Skip to main content
skill CX Codex

convex-optimization — Claude Code skill: optimization problem-solving

Skill for tackling convex optimization problems. Includes convexity verification, problem classification, and solution methods based on KKT conditions.

MIT ★ 3,951 updated 2026-01-26

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add parcadei/Continuous-Claude-v3 --skill convex-optimization -a claude-code
Open on GitHub

What convex-optimization does

The skill walks through a decision tree for solving convex optimization tasks. Start by verifying convexity: check if the Hessian is positive semidefinite and if the constraint set is an intersection of convex sets. Then classify by type—linear programming uses scipy.optimize.linprog, quadratic problems use scipy.optimize.minimize(method='SLSQP'), and semidefinite programs require CVXPY with an SDP solver.

For each type, the skill converts to standard form (minimize f(x) subject to g_i(x) ≤ 0, h_j(x) = 0) and checks KKT conditions: stationarity of the Lagrangian, primal and dual feasibility, and complementary slackness. After solving, it verifies constraint satisfaction and confirms the solution is a global minimum. The skill operates with Bash and file-reading tools, including commands like z3_solve.py prove for formal verification of optimality conditions.

What it helps with

  • Verify convexity of objective functions and constraint sets
  • Select the right solver based on problem structure
  • Convert problems to standard minimization form
  • Check KKT conditions to confirm optimality
  • Validate solutions against all constraints

Example prompts for the agent

  • Verify convexity and find the minimum for a quadratic problem using scipy.optimize.minimize
  • Solve a linear programming problem with bounds, then prove KKT conditions hold
  • Reframe a maximization task as minimization and apply interior point methods

How to install convex-optimization

1Skills CLI (Vercel)

npx skills add parcadei/Continuous-Claude-v3 --skill convex-optimization -a claude-code

2Manually: put the files in place

~/.claude/skills/convex-optimization/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/convex-optimization/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about convex-optimization

How do I install convex-optimization in Claude Code?

Run npx skills add parcadei/Continuous-Claude-v3 --skill convex-optimization -a claude-code, then read the documentation at ~/.claude/skills/convex-optimization/SKILL.md.

What does the convex-optimization skill do?

It provides a structured approach to convex optimization: verify convexity, classify the problem type (linear, quadratic, or semidefinite), convert to standard form, apply the appropriate solver, and check KKT conditions for optimality.

What tools and libraries does the skill require?

The skill uses Bash and file-reading capabilities. For actual solving, you need Python with scipy and CVXPY, plus access to z3_solve.py for formal verification of optimality conditions.

Source text

SKILL.md · MIT
---
name: convex-optimization
description: "Problem-solving strategies for convex optimization in optimization"
allowed-tools: [Bash, Read]
---

# Convex Optimization

## When to Use

Use this skill when working on convex-optimization problems in optimization.

## Decision Tree


1. **Verify Convexity**
   - Objective function: Hessian positive semidefinite?
   - Constraint set: intersection of convex sets?
   - `z3_solve.py prove "hessian_psd"`

2. **Problem Classification**
   | Type | Solver |
   |------|--------|
   | Linear Programming | `scipy.optimize.linprog` |
   | Quadratic Programming | `scipy.optimize.minimize(method='SLSQP')` |
   | General Convex | Interior point methods |
   | Semidefinite | CVXPY with SDP solver |

3. **Standard Form**
   - minimize f(x) subject to g_i(x) <= 0, h_j(x) = 0
   - Convert max to min by negating
   - Convert >= to <= by negating

4. **KKT Conditions (Necessary & Sufficient)**
   - Stationarity: grad L = 0
   - Primal feasibility: g_i(x) <= 0, h_j(x) = 0
   - Dual feasibility: lambda_i >= 0
   - Complementary slackness: lambda_i * g_i(x) = 0
   - `z3_solve.py prove "kkt_conditions"`

5. **Solve and Verify**
   - `scipy.optimize.minimize(f, x0, constraints=cons)`
   - Check constraint satisfaction
   - Verify solution is global minimum (convex guarantees this)


## Tool Commands

### Scipy_Linprog
```bash
uv run python -c "from scipy.optimize import linprog; res = linprog([-1, -2], A_ub=[[1, 1], [2, 1]], b_ub=[4, 5]); print('Optimal:', -res.fun, 'at x=', res.x)"
```

### Scipy_Minimize
```bash
uv run python -c "from scipy.optimize import minimize; res = minimize(lambda x: (x[0]-1)**2 + (x[1]-2)**2, [0, 0]); print('Minimum at', res.x)"
```

### Z3_Kkt
```bash
uv run python -m runtime.harness scripts/z3_solve.py prove "kkt_conditions"
```

## Key Techniques

*From indexed textbooks:*

- [Additional Exercises for Convex Optimization (with] Finally, there are lots of methods that will do better than this, usually by taking this as a starting point and ‘polishing’ the result after that. Several of these have been shown to give fairly reliable, if modest, improvements. You were not required to implement any of these methods.
- [Additional Exercises for Convex Optimization (with] K { X = x Ax yi } where e is the p-dimensional vector of ones. This is a polyhedron and thus a convex set. Rm has the form − The residual Aˆx − Describe a heuristic method for approximately solving this problem, using convex optimization.
- [Additional Exercises for Convex Optimization (with] We then pick a small positive number , and a vector c cT x minimize subject to fi(x) 0, hi(x) = 0, f0(x) ≤ p + . There are dierent strategies for choosing c in these experiments. The simplest is to choose the c’s randomly; another method is to choose c to have the form ei, for i = 1, .
- [Additional Exercises for Convex Optimization (with] We formulate the solution as the following bi-criterion optimization problem: (J ch, T ther) cmax, cmin, 0, minimize subject to c(t) c(t) a(k) ≤ ≥ t = 1, . T The key to this problem is to recognize that the objective T ther is quasiconvex. The problem as stated is convex for xed values of T ther.
- [nonlinear programming_tif] Optimization Over a Convex Set** - Focuses on optimization problems constrained within a convex set. Optimality Conditions:** Similar to unconstrained optimization, but within the context of convex sets. Feasible Directions and Conditional Gradient** - Explores methods that ensure feasibility within constraints.

## Cognitive Tools Reference

See `.claude/skills/math-mode/SKILL.md` for full tool documentation.

© parcadei · MIT · parcadei/Continuous-Claude-v3

Similar skills

All skills for AI and ML →

From the same set161

The whole set →
root-finding root-finding is a Claude Code skill that selects a numerical method for finding equation or system roots based on available derivatives and a known interval. CX Codex +1 ★ 3,951repo stars banach-spaces Tackles problems in functional analysis involving Banach spaces. Includes strategies for proving key theorems: completeness, Hahn-Banach, open mapping, closed graph, and uniform boundedness principle. CX Codex +1 ★ 3,951repo stars contour-integrals contour-integrals — a Claude Code skill for solving contour integral problems in complex analysis. It guides you through choosing the right contour, locating poles, and applying the residue theorem to evaluate definite integrals. CX Codex +1 ★ 3,951repo stars arbiter arbiter runs unit and integration tests, investigates failures, and checks an implementation against acceptance criteria. Saves its report in the project. CX Codex ★ 3,951repo stars connectedness A skill for Claude Code with strategies for connectedness problems in topology. CX Codex +1 ★ 3,951repo stars analytic-functions analytic-functions is a Claude Code skill for checking analyticity, verifying Cauchy-Riemann equations, and computing power series and analytic continuations. It activates for theory of functions questions. CX Codex +1 ★ 3,951repo stars math-intuition-builder math-intuition-builder develops mathematical understanding by guiding you through concrete examples, diagrams, and analogies before diving into calculation. Useful when you need solid intuition before solving a problem. CX Codex +1 ★ 3,951repo stars constrained-optimization Guides Claude Code through constrained optimization problems. It classifies constraints, formulates Lagrangians, and checks KKT optimality conditions. CX Codex +1 ★ 3,951repo stars

Project sponsors

Companies that keep this analytics open