When creating a new repository on GitHub you can provide instructions for a GitHub Copilot coding agent to execute. The agent picks up those instructions, does the work in the cloud, and opens a pull request with the results — all from a single prompt at repository creation time.
You can use this to create a C# BenchmarkDotNet project. The prompt can include the specific benchmarks to run and everything needed to get benchmark results in a pull request. The agent creates the project, builds in release mode, runs the benchmarks in the cloud, and opens a pull request.
This is a useful technique for benchmarking — you describe the scenarios and configuration upfront, and the agent handles project creation, building in release mode, and running the benchmarks. Writing a detailed prompt means you get benchmark results and a reviewable project without any manual setup.
Prompt
Here’s a prompt to use as a starting point:
Create a C# BenchmarkDotNet project comparing <your scenarios here>.
## Setup
- <provide any specific instructions needed for setup>
## Benchmark building instructions
- Create the C# benchmarking code.
- Include allocations in the benchmarking.
- Ensure the benchmarking is built for reliable results.
- Release build
- Anything else necessary for reliable results
- Run benchmarking in the cloud agent.
## README.md specific instructions
- Include benchmarking results as markdown in README.md.
- Include any caveats to benchmarking in the cloud agent in README.md.
- Include this entire exact prompt in README.md as a code block, make it somewhere below the benchmarking results and any caveats. Use quadruple backticks on the outermost code block since this prompt contains triple backtick code blocks.
- Update the README.md title to be "<your title here>".
- Include basic repository license information at the end of README.md.
## Benchmark: <Name>
```csharp
// Setup
<your setup code>
// Benchmark
<your benchmark code>
```
## Benchmark: <Name>
```csharp
// Setup
<your setup code>
// Benchmark
<your benchmark code>
```
(repeat for each scenario you want to benchmark)