2026-05-14

GitLink

Today I tried to push forward, but progress was slow while interacting with the Agent. I realized a problem: if I don’t have enough cognitive load of the current project’s information, I lack confidence to move ahead—because I’m not sure if the AI is making things up.

I discovered another issue: analysis might come with performance costs. To evaluate the health of an open-source project—such as the interaction between developers and contributors—I need to retrieve PR information from the repository. This includes not only the metadata I need (various numerical metrics) but also the full Git diff content. Some repositories may have hundreds or thousands of PRs. It’s completely impractical to pull down every PR for each evaluation. I think a better solution would be to have a directly queryable database—if GitLink’s database could expose a direct query API, that would be ideal.

Tomorrow’s task: first, confirm exactly what metadata can be obtained via gitlink-cli and what it can support. I had the Agent do this task before, but I didn’t verify it in time…

Regarding the performance cost issue, I believe the key in the competition is to demonstrate my Agent skill design ability, not to solve the engineering problem of performance cost. Once the skill is well designed, it can achieve the same effect in any environment.

Tasks for Tomorrow

• [ ] Determine which data can be obtained
• [ ] Determine which aggregate metrics can be calculated
• [ ] Advanced tasks
  • [ ] Think about the health workflow