How to get better results for complex problems in Cursor

1// In the Cursor Chat panel:
2// 1. Click the model dropdown at the top
3// 2. Select one of these for complex problems:
4//    - Claude 3.7 Sonnet (with thinking) — 2x credits
5//    - o3 — deep reasoning, 2x credits
6//    - Gemini 2.5 Pro — large context, good for math
7// 3. Toggle MAX mode for extended context and 200 tool calls
8//
9// Thinking models use step-by-step reasoning before
10// generating code, dramatically improving correctness
11// for DP, graph, and math problems.

1// Cursor Chat prompt (Cmd+L):
2// Problem: Given a weighted directed graph, find the shortest
3// path from source to all other vertices. Handle negative
4// edge weights but detect negative cycles.
5//
6// Input: adjacency list as Map<string, Array<{to: string, weight: number}>>
7// Output: Map<string, number> (vertex -> shortest distance)
8// Constraints:
9// - Up to 10,000 vertices and 50,000 edges
10// - Edge weights can be negative
11// - Must detect negative cycles and throw an error
12// - Target complexity: O(V * E)
13//
14// Edge cases:
15// - Disconnected vertices (return Infinity)
16// - Self-loops with negative weight
17// - Empty graph
18//
19// Implement Bellman-Ford algorithm in TypeScript.

1// Add to your prompt:
2// Test cases:
3// 1. Simple graph: A->B(1), B->C(2), A->C(5)
4//    shortest from A: {A: 0, B: 1, C: 3}
5//
6// 2. Negative edge: A->B(4), B->C(-2), A->C(5)
7//    shortest from A: {A: 0, B: 4, C: 2}
8//
9// 3. Negative cycle: A->B(1), B->C(-1), C->A(-1)
10//    should throw Error('Negative cycle detected')
11//
12// 4. Disconnected: A->B(1), C (no edges)
13//    shortest from A: {A: 0, B: 1, C: Infinity}
14//
15// Verify your solution passes all test cases.

1// Cursor Chat prompt (Cmd+L):
2// Before writing any code, explain your approach:
3// 1. Which algorithm will you use and why?
4// 2. What are the key data structures needed?
5// 3. How will you handle edge cases?
6// 4. What is the expected time and space complexity?
7// 5. Are there any tricky implementation details?
8//
9// After I approve the approach, implement the solution.

1// Follow-up prompt if a test case fails:
2// Test case 3 (negative cycle) does not throw an error.
3// The Bellman-Ford algorithm needs one more iteration
4// after V-1 rounds to detect negative cycles. Fix only
5// the negative cycle detection logic. Do not rewrite
6// the entire function.