Proposal – Context & execution observability for more reliable Cursor agents

Showcase
,
1

Feature request for product/service

Indexing

Describe the request

Hello Cursor team,

I’m reaching out to share a project I developed from a real usage need while working intensively with Cursor.

In longer sessions, I observed cases of context drift and pollution: irrelevant file reads, noisy accumulated state, and occasional loss of precision. To diagnose this more precisely, I built Context Vision — a system that reconstructs session state step by step and traces what the agent reads, writes, and uses.

I then started a second project focused on execution observability. My goal is to combine both into a broader observability layer for complex agent workflows.

I believe this could help Cursor:

  • identify unexpected agent behaviors earlier
  • improve context construction quality
  • guide agents to focus on relevant project areas
  • reduce pollution from weakly relevant files/context
  • provide measurable signals for continuous product improvement

I’m not contacting you primarily for funding.
My main request is a professional exchange: feedback, integration/migration guidance, or redirection to the right person (PM / Engineering / DevRel).

If helpful, I can share a short demo, real cases, and a concise technical brief.

Thank you for your time and for building Cursor.

Best regards,
tayier
[email protected]
https://www.linkedin.com/in/tayier-dev-ai-data/?locale=fr-FR

5

Hey @Snorbik ,

For tool/integration pitches, [email protected] is the right place. A forum post won’t reach the people who evaluate those.

On the technical side, this thread on richer agent transcripts / observability is the better place to dig into gaps - there’s already an active back-and-forth there. Hooks and the headless CLI’s --output-format stream-json cover a lot of the lifecycle/event surface today; step-by-step session reconstruction is the piece that isn’t there yet.

We’ll be tracking this thread to gauge interest from the community.