Product feedback for Cline

Cline reads files. Export a CobaltCapture review as .md, drop it in your workspace, and Cline acts on it directly.

This is the playbook for getting visual product feedback into a Cline session without losing context between captures and code. For the same workflow applied to other tools, see the full feedback for AI coding agents hub.

The problem with feedback in Cline workflows

Cline sits inside VS Code with full file-system access. It reads, writes, and runs commands across your workspace. That capability is wasted when the feedback it's working from is a single screenshot pasted into the chat panel with one line of context underneath.

The default flow is fragmented. Someone reviews the staging build. The screenshots land in a Downloads folder. The descriptions land in Slack. The source URLs land nowhere. By the time you sit down with Cline, you're either dragging images into the chat one at a time or summarizing six findings into a single paragraph and hoping Cline picks the right component to fix.

The structure that makes Cline good at code, file-aware, multi-step, persistent across the workspace, is the same structure feedback should arrive in. A markdown file that lives in the repo, that Cline can read alongside your source.

The CobaltCapture workflow with Cline

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 part. 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. From the root of your project:

curl -fsSL https://cobaltcapture.com/r/<slug>/markdown > feedback.md

Now feedback.md sits in your workspace alongside the code Cline is about to touch. Open Cline in VS Code and point it at the file:

Read feedback.md and work through the items in order. For each finding,
summarize what you understood, propose the fix, then apply it.

Cline reads the markdown like any other source artifact in the project, follows the embedded screenshot URLs when it needs visual context, and works through the findings against the actual components you captured.

Example prompt

I just published a review of the checkout flow at feedback.md in the
project root. Please read it, then address items 1, 2, and 3 in order.

For each item:
1. Summarize the finding in your own words so I can confirm.
2. Identify the component or file you'll change.
3. Apply the fix and run the test suite.
4. Pause for my confirmation before moving to the next item.

Why this works for Cline

Cline is file-aware by design. The cleanest input is a file in the workspace, not a payload in the chat. Once feedback.md is in the project root, Cline treats it the same way it treats README.md or any source file, addressable, re-readable, and discoverable across sessions.

The embedded image URLs in the markdown render when Cline opens the file in VS Code's preview, and Cline can fetch them when it needs the visual context to disambiguate a finding. The dictated commentary underneath each screenshot carries the nuance a one-line caption can't. "This happens on iOS but not Android, only after the keyboard dismisses", which is the difference between a shallow patch and a real one.

The .md also persists across Cline sessions. When the context window resets or you close VS Code and come back tomorrow, the feedback is still in the repo. You don't re-paste anything.

Alternatives and tradeoffs

You could paste screenshots straight into Cline's chat. It works for one finding. Past three, the chat fills with disconnected images, the descriptions lose their anchor, and the whole thing evaporates when the session ends.

You could file Linear tickets for each finding and connect Cline to Linear via an MCP integration. That works once feedback has been triaged into proper tickets. It's overkill for the "I just walked through staging and found six things" stage, which is exactly where CobaltCapture lives.

You could write the markdown by hand, screenshots, captions, source URLs, commentary, formatted manually. It works. It's just slower than capturing and dictating. CobaltCapture exists so the artifact is structured the same way every time, against the same agent-feedback workflow, regardless of who ran the review.

Get started

CobaltCapture is free to try. Capture your first review, save the markdown export into your workspace, and let Cline read it on the next session.

Capture your first review.

About a minute from open tab to a shareable URL your agent can ingest.

Start capturing