Product feedback for Cursor

This page is for anyone shipping with Cursor who wants visual feedback the agent can actually act on. It walks through the workflow and a paste-ready prompt pattern. See the rest of the feedback for AI coding agents hub for the same playbook applied to other tools.

The problem with feedback in Cursor workflows

Cursor's composer is good at code. It is not good at guessing what's on your screen. The default flow is brittle: take a screenshot, drag it into the chat, type a sentence about what's wrong, hope the model has enough context. With one bug it works. With five findings on a staging build, the composer fills up with disconnected images, your descriptions lose track of which screenshot they belong to, and Cursor produces patches against the wrong component.

The structured alternative is a markdown document where every screenshot has a caption next to it, every caption names a real selector or component, and the whole thing fits in a single composer paste.

The CobaltCapture workflow with Cursor

Open the staging URL you want feedback on. Open cobaltcapture.com in a new tab and hit Capture screen. Pick the staging window. Drag a box around the broken component, then hit Dictate and talk through the problem out loud. "The submit button overflows the container at widths under 380px on iOS Safari." Repeat for each finding. Hit Publish.

You get a public URL like cobaltcapture.com/r/<slug> and a one-click markdown export. In Cursor:

@feedback.md please fix items 1-3 in order

Or if you'd rather paste the URL directly into the composer:

The staging review for the checkout flow is documented here:
https://cobaltcapture.com/r/abc12345

Please address the three findings in order. Show me the diff for each
before moving on.

Cursor follows the URL, pulls the screenshots inline, and produces patches against the actual components named in the captions.

Why this works for Cursor

Three reasons.

First, Cursor's composer is markdown-native, pasted markdown renders correctly and embedded images are processed inline as visual context, not lost as attachments. Second, the @file reference syntax means a downloaded .md export in your repo becomes addressable across sessions; once feedback.md is in the project root, you can reference it from any Cursor chat. Third, dictation captures the kind of context a typed bug report skips. "This happens on iOS but not Android, only after the user has scrolled past the fold." That nuance shapes the patch.

Alternatives and tradeoffs

You could keep screenshotting and typing into the composer. That's fine for one or two findings. Past five it falls apart because the screenshots stop being labeled and your text loses its anchor.

You could record a Loom and transcribe later. Cursor cannot watch the video, and the act of transcribing makes Loom the worst of both worlds: time spent recording plus time spent typing.

You could file Jira tickets and link Cursor to the Jira API. That works if your team's primary feedback artifact is already a Jira ticket. CobaltCapture is for the messier reality where the feedback exists but no one's filed a ticket yet, and the next step is "ask the agent to fix it." For the canonical pattern behind it, see screen capture to markdown.