The Cobalt Capture blog

A reusable prompt structure for handing Windsurf a cropped still, one numbered pin, and three tight bullets, so it edits the right element on the first turn.

A short vocabulary sheet for FAQ authors: question, trigger, screen state, and resolution, so every entry you capture reads the same way.

Your corporate laptop blocks Greenshot and every other download, but you still have to report a bug today. Here is why installs get blocked and what to use instead.

Half of what people use ShareX for is capture, annotate, upload, and share a link. Here is which of those jobs run in a plain browser tab, and which still belong in ShareX.

The Lightshot reflexes you built over years (free-floating arrows, dumping to prnt.sc, one shot per bug) work against you in a browser capture tool. Here is what to unlearn.

A hosted image is fast and often enough. A captured still with a comment attached and reply threads on one link is a different job. Here is how to pick.

Before you send a product review to a client, run its markdown through Perplexity to catch a shaky claim. Here is the actual input and the actual result.

A flow walkthrough and an onboarding pass both mean walking a screen and marking problems, but they answer different questions. Here is which one you are running.

Three practices for capturing hero, subhead, and proof sections so a copywriter opens your review and starts rewriting, not decoding.

Capture a pricing page in pieces, pin the parts that confuse buyers, and publish one link the founder opens on a phone in two minutes.

A step-by-step doc and a short published review pull in different directions on a one-off handoff. Compare them on effort, edits, and what the receiver does next.

A reusable markdown bug report template built around a cropped still, numbered pins, and observed versus expected. Copy it, fill it, and hand Cline enough to fix in one turn.

You captured a rival's checkout and pricing page. Now what? Answers on where those screenshots live, who can read the link, and how long an anonymous teardown sticks around.

A screen capture that comes back black or empty almost always has one of three causes. Here is how to spot which one you hit and fix it in seconds.

A step-by-step way to review a Figma prototype in the browser and hand off notes a developer can act on, even on a file you can only view.

Walk from a captured finding to a Linear issue that has the screenshot, the pinned spot, and the repro steps already in the body, so the developer starts fixing instead of asking questions.

Turn a captured, pinned screen and a spoken repro into a Jira ticket body a developer can act on, no per-seat connector required.

Walk from a captured review to a GitHub issue your team can triage, with the screenshots visible right in the issue body and a link back to the full context.

A step-by-step way to move a visual bug report into a GitHub issue with the screenshot embedded, using clean markdown and no paid add-on.

Capture the broken screen, publish, then have Cursor read the whole review over MCP. The context for the fix is already in front of it before you type a prompt.

You sent the staging link days ago and still no notes. The problem is usually the tool you asked the client to log into. Here is a faster loop.

A step-by-step way to package one UI bug as markdown Claude Code reads straight through, screenshot included, so it lands the fix on the first attempt.

Click through your build, capture each rough spot in order, and publish one review the team acts on instead of chasing scattered Slack notes.

Vague feedback makes an agent guess or stall. Here are the five fields that let an LLM act on your report in a single pass, and how to capture them fast.

The change looks right in Figma but wrong in the build. Here is how to review the built screen and hand a developer feedback they can act on without a meeting.

When you paste a feedback link into your coding agent, the image often falls away and the agent works from a URL it cannot see. Here is why that happens and how to hand it the actual screen instead.

Found a bug during a test pass? Here is the fastest route from broken screen to a filed report your developer can act on, all in one browser tab.

Wire a published Cobalt review straight into your coding agent. Here is the one-command MCP setup for Claude Code, Cursor, and Codex, and the tools each one gets.

Four common jobs, four different right answers. Match the tool to whether you need a fast handoff, developer bug repro, client sign-off, or comments on a live page.

Running the study against localhost skips the bot-blockers you hit on production and gives the agent repo access to explain why a flow breaks, not just that it does.

People confuse the tracker that holds the ticket with the tool that makes the ticket reproducible. They solve different problems. Here is how to pick.

An agent's chat summary vanishes when you close the tab. A captured, annotated review stays. Here is where each one fits and what each catches.

You just got blocked from installing a screenshot extension. Here is how to capture, mark up, and share a screenshot from one browser tab instead.

A step-by-step method for having your coding agent walk your product like a new user and hand you the findings as a review link, PDF, or markdown.

A dedicated SOP platform and a single browser capture solve different problems. Here is how to tell which one your process actually needs.

Cropping and pinning rules for capturing a step on a crowded screen, so your quick doc still makes sense to the person reading it three months from now.

An FAQ entry that made sense the day you wrote it can fall apart the second time someone reads it. Two capture mistakes cause most of that damage, and both take seconds to fix.

Padding is not margin. Disabled is not read-only. A short vocabulary list for spacing, hierarchy, and state so a comment means the same thing to the person who wrote it and the person who fixes it.

Instead of typing a dry status update, capture each screen and talk through what you see. The client hears why, and the review shares as a link, PDF, or Word doc.

Read the analytics page on a real published review and see what visit counts and visitor geography tell you about which stakeholders actually opened the link.

Asking a busy stakeholder to install something on their phone is where feedback dies. A plain review link that opens in any mobile browser sidesteps the whole problem.

Your designer is on a 27-inch monitor. Your client is thumbing through on an iPhone in a taxi. Here is why invite-based review tools stall on that split, and how a public link fixes it.

Copy this three-part structure into one review: current state, intended state, and constraints. Each part becomes an item an agent can read and act on.

An accessibility finding a developer cannot locate never gets fixed. Here are the mistakes that make a11y feedback hard to act on, and how to correct each one.

You send a client a review link and they hesitate. Here are the questions they actually ask, answered plainly, so they open it and start reviewing.

A five-step workflow for cropping the exact subcomponent, dictating a single change, and sharing a markdown link v0 can read.

Dictation is faster for long reproduction steps but wrong for exact values. Here is when to talk through a bug, when to type it, and how to handle Firefox.

Region, element, and state words that keep their meaning when your feedback reaches Claude Code as markdown. Use these instead of vague pointing.

Watch three FAQ entries get built in a single sitting with cropped screenshots, dictated answers, and numbered pins, then published to one link.

How to place numbered pins on a single screenshot so a developer reads your bug report as an ordered checklist, not a scavenger hunt.

A step-by-step workflow: capture a screen, say what's wrong, publish, and hand Cursor a markdown URL it can read. All in one browser tab, in under a minute.

You captured a review of a broken screen. Here is how to hand it to Claude Code so each item becomes a discrete task it can work through.

A step-by-step workflow for pulling key moments from recorded interviews into a single research review, then sharing one link the whole team can read.

You send a review tool to a client or stakeholder and it dies at the install screen. Here is why external reviewers refuse to set up an account, and how to route around it.

Teams switching from Jam.dev to a link-based review make the same handful of errors. Here is what stalls triage and how to correct each one.

A three-part item structure for product reviews that a busy PM can read and act on in two minutes: one screenshot, one observation, one suggested next step.

You published a review without an account. Here is exactly how long it sticks around, what claiming does, and what happens at day 30.

Say "the primary button in the modal header" instead of "that thing at the top." A working vocabulary for naming buttons, labels, and regions in feedback.

Three defensible rules for framing a cropped still so an agent identifies the right component, plus what goes wrong when you ignore each one.

Six steps your client can run on the staging site in about ten minutes, capturing stills and dictating notes into one link your developers can act on.

Walk a competitor's checkout, capture each screen, dictate what you notice, and end with one link your team can read, comment on, and copy from.

Pastel organizes client feedback into review rounds. A per-build link keeps each version separate and skips the login. Here is how they differ and who each suits.

Capture the built screens, comment on each one, publish, and paste the markdown export into Cursor as a task list. Here is the exact order to do it in.

Vague selectors, no screenshot, five bugs in one message: the patterns that make Bolt guess wrong, and the concrete fix for each.

The client cannot reach your staging URL because of basic auth, VPN, or an IP allowlist. Capture the screens instead and send a link they can open.

Common access questions about a published review link: who can read it, who can post comments on items, and who can mark them resolved.

Copy this four-step sequence and fill it in as you walk your own signup flow. One captured still and one clear note per screen, ending in a link the team can act on.

Agent-readable feedback is a defined term. Here is what it means in practice, with a short example of a cropped still and the markdown that comes with it.

The crop rectangle decides which element an agent touches. Four defensible rules for what to keep, what to cut, and the failure mode when you get it wrong.

Walk through capturing a Lovable preview, pinning the exact spot, and pasting the markdown into the chat so the fix lands on the first try.

Assemble captured screens from session replays plus spoken observations into a single review, then send your PM one link they can read, resolve, and export.

An agent will happily edit the wrong element if your markdown bug report is vague. Here are the specific mistakes that cause it and the concrete fix for each.

Aider works from the terminal and reads pasted text well. Here is a small, reusable template that pairs one cropped still with three bullets so the change lands right.

A worked example: the same bug report shown as a published review link for people and as clean markdown for Cline to read and fix.

The Replit Agent built the screen, the preview is broken, and you need to say exactly what and where. Here is the capture-comment-return loop, step by step.

Capture a rendered page, crop the broken region, and feed the markdown to Zed AI. Here is the exact input and the exact result, step by step.

A pinned screenshot and one dictated sentence beat a paragraph of description when you prompt Windsurf. Here is the step-by-step flow.

A step-by-step way to send v0 a precise visual fix so it edits the component you already have instead of regenerating the whole screen.

Capture your Bolt preview, mark up what needs fixing, and hand Bolt a single markdown URL it can read. Here is the exact order of steps.

An agent can only act on what your screenshot points at. Here is how to crop, pin, and word a comment so the change lands on the intended element.

Follow one recorded interview from playback to a published review link. Real captures, real dictated notes, and the exact output your team gets at the end.

Markup anchored to a live page drifts or vanishes after a deploy. A cropped still frame records exactly what you saw, so your feedback still makes sense next week.

Answers to the real questions about publishing a review without signing in: the 30-day window, magic link sign-in, and how to keep the review for good.

A step-by-step walkthrough of publishing a review, collecting comments, and marking each one resolved, covering the lightweight tracking most people reach for a bug tracker to do.

The analytics page on a published review shows visit counts and visitor geography. Here is how to read those numbers before a launch, and what they can and cannot tell you.

Firefox does not support voice dictation, but a reviewer there can still capture a screen, type comments, pin the spot, and publish a link. Here is exactly what changes and what does not.

Typing slows a review pass to a crawl and strips out context. Talking through each captured screen keeps you moving and preserves the why behind every note.

Frame, crop, pin, item, review link. Define the words you use to describe a screen so your prompt to an agent or a teammate points at the exact spot you mean.

Go from a broken screen to a markdown link Aider can read in about two minutes. Here is the exact sequence, step by step.

A comment like "the button is confusing" makes the receiver guess which button. Numbered pins point at the exact control, so feedback lands where you meant it.

The questions a first-time reviewer actually asks, answered plainly: capturing a screen, adding pins and comments, publishing a link, and picking PDF, Word, or markdown.

Cursor can only act on what you actually tell it. Here are the feedback mistakes that send it guessing, and how to correct each one.

Take one messy note about a checkout bug and rewrite it into structured feedback an LLM can act on. The actual inputs and the actual markdown output, shown in full.

You captured a bug. Do you send a still frame with notes or a screen recording? Here is how each one lands when an AI coding agent has to act on it.

A precise glossary of markdown screenshots: what the term means, what a review item and a numbered pin are, and why agents read the format cleanly.

Vague feedback wastes an agent's time. Here is how to capture a screen, name the problem, and hand over text the agent can turn into a fix.

A step-by-step accessibility pass that ends in a shareable doc: capture the problem screen, name the WCAG issue, point at the spot, and publish.

A code diff tells you what changed, not how it looks. Here are five specific practices for adding screenshots to a pull request review so the author can act on every comment.

Feedback lands in five places at once: email, Slack, a forwarded screenshot, a phone call. Here is why it scatters and how a single review link fixes it without making anyone learn software.

A short glossary for reviewers whose feedback ends up in an agent's context window. Define the artifact, the pin, the prompt, and what each one changes.

A step-by-step procedure for reviewing a staging build and sending issues someone (or an agent) can fix without a follow-up call.

A concrete walkthrough from opening a tab to exporting markdown. No install, no extension, no signup. The output reads cleanly for people and for agents.

A field-by-field QA bug report template you can copy into any review. Covers what to include, when to trim it, and how to attach the screenshot.

Capture once, share twice. Walk through the exact steps to send a client the polished review link while your developer gets the same notes as markdown.

Walk through a design review from opening a browser tab to sending a public link. Each step has a clear outcome the team can act on.

A recording shows everything and proves nothing scannable. A still capture forces you to say what matters. Here is how to choose between them for written feedback.

Agent feedback is not a vibe or a workflow buzzword. It is a specific artifact with a specific shape. Here is what it contains and why each piece is there.

Walk the build, narrate what you see, publish a link. A step-by-step workflow for sending a distributed team feedback that stands on its own without a call.

Your client is asleep when you ship, and you are asleep when they wake up. Here is how to give feedback that does not need a shared calendar slot.

You opened Cobalt Capture, saw the Capture screen button, and paused. Here are the questions first-time reviewers actually ask, answered plainly.

You built a screen, it is almost right, and you want an agent to finish it. Here is the handoff that gets the agent looking at the same pixels you are.

A short, screenshot-led how-to template your team can reuse. Each part explained, with notes on when to trim, expand, or rewrite a section.

BugHerd's pinned-on-page bug tracker shines on long projects. A no-login review link wins when you just need feedback someone can act on today.

You need to hand off a process to one teammate by Friday. You don't need an SOP platform for that. Here's a 20-minute method using a browser tab.

Vibecoding stalls when the loop between seeing a problem and telling the agent about it is sloppy. Here is what breaks and how to fix it.

Capture the exact screens a user sees, narrate each step, and ship a complete FAQ entry to your help center without opening a docs tool.

Open a tab, capture a screen, talk through what you see, publish, and walk away with a markdown file a developer or coding agent can read straight away.

Vague comments, missing screenshots, mixed priorities. The specific feedback mistakes that cost a round trip, with the concrete correction for each.

Claude Code will happily act on bad input. Here are five specific ways UI feedback goes wrong before it reaches the agent, and what to send instead.

Walk a signup flow as a first-time user, capture each friction point in order, and hand the team a review link they can act on the same day.

Snagit is a great desktop capture suite. For review feedback that someone else has to act on, a publish-and-share link often gets there faster.

Walk through a competitor's signup, pricing, and core flow in one sitting, capture each screen with commentary, and hand the team a single link they can open.

A procedural walkthrough: capture the screen, write the comment, pin the spot, export markdown, and hand your agent a prompt it can actually execute.

Coding agents cannot watch your Loom. They need text with structure. Here is why video breaks the handoff and what actually lands as actionable input.

Diffs hide the things that actually matter in UI work: spacing, hover states, focus rings, mobile layout. Here is how to review a PR on screen and attach the result.

Why client feedback arrives as Slack screenshots, forwarded emails, and texts saying the logo looks weird, and how a single no-signup link pulls it back into one place.

A practical walkthrough for reviewing the built product against the design: cropped stills, spoken comments, numbered pins, and a shareable link the team can act on.

A concrete pre-launch staging review walkthrough: what to check, how to capture it, and how to hand it off so the fixes actually ship before go-live.

Markup.io organizes long client review cycles around projects and versions. A quick review link skips all that. Here is which fits which job.

CleanShot X is a beautiful Mac annotation app. A browser-based review link skips the install and gets feedback to the receiver faster. Which one fits the job?

Snagit is a powerful desktop capture app. For one-off reviews you need to share with someone, a browser tab often gets the job done faster.

Loom is great for talking heads and walkthroughs. For most feedback, a video forces the receiver to scrub, transcribe, and guess what to fix.

BugHerd gives agencies a real bug tracker on top of every client site. Sometimes that's the right tool. Sometimes a single review link is the better fit.

Scribe is built for SOP libraries. For a one-time how-to you need to send a colleague today, a browser capture often fits the job better.

IT-blocked extensions stall reviews for weeks. Here's why approval is slow, and a browser-only way to capture and share screen feedback in the meantime.

The precise vocabulary for bug reports a coding agent can fix without follow-up questions. Each term defined, with practical notes on when it matters.

Loom is great for talking through a screen. It is bad for handing work to a coding agent. Here is how a recorded video stacks up against a markdown review on the axes that matter.