Bugbot Autofix: Your Pull Request Reviewer That Actually Fixes Things

Here’s the thing about code reviews: someone spots an issue, flags it, and then you’ve got to context-switch back, hunt down the problem, and fix it yourself. That’s at least three separate tasks when really it should be one.

Bugbot can now automatically fix issues it finds in pull requests. You’re not getting a list of problems with commentary anymore; you’re getting pull requests that are genuinely ready to merge (or at least, further along than they were). This matters for a few reasons:

• Your team spends less time in the back-and-forth loop. Actual momentum on the work.
• Project managers can see progress moving without waiting for the seventh round of “tiny fixes.”
• Developers get to focus on the logic, not playing patch-and-check.

If you’re already running Bugbot for catching issues, this is a no-brainer upgrade. If you’re not, it’s worth trying on a smaller branch just to see how many cycles it saves you.

CLI Gets Mermaid ASCII Diagrams and Keyboard Shortcuts

Working in the terminal and suddenly needing to visualise something? Yeah, that used to mean either opening another window or squinting at raw Mermaid code. Not anymore.

The CLI now renders Mermaid diagrams directly as ASCII art, which sounds niche until you’re actually mid-flow and trying to explain a data architecture to a teammate over a quick screen share. Flowcharts, sequence diagrams, state machines, class diagrams, and ER diagrams all render inline. Press Ctrl+O and you can flip between the rendered diagram and the raw code without breaking your terminal session.

They’ve also added some basic keyboard shortcuts that make the plan mode feel less clunky:

• Arrow keys to navigate options
• Enter to execute the selected option
• Shift+Enter to hand off plans to the cloud straight away

Typing /plan brings you back to your current plan and its menu whenever you need it. It’s the kind of quality-of-life stuff that doesn’t sound revolutionary until you realise you’re wasting two minutes every time you need to jump back to a decision menu. Multiply that by a dozen times a day and it adds up.

Cloud Handoff for Plans: Keep Your Workflow Unbroken

The CLI now lets you hand off plans to the cloud for execution instead of running everything locally. Why does this matter? Because sometimes your local machine is already chewing through something else, or you’re on a shoddy connection, or you just want to spin up a heavier compute job without crushing your laptop’s battery.

You generate a plan in the terminal, see what it’s suggesting, and then decide: build locally or push it to the cloud. The persistent menu stays visible, so you’re never hunting for where you left that decision point.

Expanded Agent Capabilities: Running More in Parallel

Cursor 2.0 lets you spin up to eight parallel agents running in isolated cloud environments simultaneously. Before, you’d essentially queue tasks; now you’re genuinely running them side by side.

This translates to a few practical scenarios:

• Frontend architect agent and backend database agent working on separate parts of the same feature at once
• One agent writing tests while another refactors code
• Parallel work on infrastructure, API design, and migration tasks without stepping on each other’s toes

The agents run in Ubuntu VMs with internet access and use Git worktrees to keep code isolated. Most tasks wrap up in under thirty seconds. They can even open pull requests when they’re done and run self-tests through a native browser tool.

For teams scaling up their use of agents, this is where the rubber hits the road. You’re not just automating individual tasks; you’re genuinely orchestrating parallel workflows.

Increased Usage Limits for Auto and Composer

If you’ve been using Cursor’s agent features, you might’ve hit usage limits faster than expected. Cursor’s now split usage into two pools:

• Auto + Composer: Significantly higher limits (Composer 1.5 has three times the limit of Composer 1, and through mid-February that was bumped to six times)
• API: Priced separately; unchanged from before

This basically means you can run agents more liberally without constantly hitting walls. Individual plans now get at least 20 dollars’ worth of API usage monthly, with more on higher tiers. The practical upshot is developers can experiment with agentic workflows without penny-pinching every request.

Cursor 2.5 Adds Plugin Marketplace and Better Sandbox Controls

Cursor’s launched a plugin marketplace with an installer, so you can extend capabilities without wrestling with manual setup. You can grab plugins for design, databases, payments, and analytics. That’s the kind of thing that doesn’t sound huge until you realise you’re not copying config files or hunting documentation anymore.

Alongside that, there’s granular sandbox network and filesystem controls. If you’re running code in isolated environments and want fine-grained policy control over what gets network access and what touches the file system, you’ve now got that knob to turn.

Subagents can now run asynchronously and spawn other subagents, which means complex workflows actually nest properly instead of bottlenecking at each step.

The Practical Bottom Line

These updates aren’t about flashy AI theatrics. They’re about removing friction: fewer round-trips on reviews, better terminal workflows, genuine parallelism for agents, and more control over how your tools run. If you’ve been considering Cursor or already using it, these are the kinds of refinements that compound over time. A minute saved here, a context-switch avoided there, and suddenly your whole day feels less fragmented.

Worth trying on a small project to suss out which bits actually fit your flow. Different teams will get different mileage depending on how heavy they’re leaning into agents versus traditional coding.