Everything is up to date. Seems Cursor can sometimes read my code but can’t edit anything. I’ve retried dozens of time over the last 2 days.
“Still blocked right now — the execution backend is down again on this turn (Execution backend unavailable even on file read), so I can’t apply or verify edits yet.”
I’ve tried all the usual stuff like
restarting WSL
reinstalling the Anysphere extension
deleting my ~/.cursor-server
removing unnecessary extensions
Don’t matter what project/folder I try to work with, I get the same issue. Also doesn’t matter what AI model I pick.
Steps to Reproduce
Anything I ask the agent to do produces this type of response.
Hi Gary (and Tianhao) — sorry for the rough couple of days, and thanks for the detailed report. You’re not doing anything wrong: this is a known issue we’re actively working on. On Windows + WSL2 the extension host that runs the agent’s tool calls can become unresponsive, so every tool call (Read, Grep, shell) comes back as “Execution backend unavailable.” We’ve confirmed an uptick matching your timeframe, and the team is on it.
A few things worth trying in the meantime:
Fully quit and relaunch Cursor (more reliable than Reload Window) to clear a stale extension-host session.
Change folder scope — this is sensitive to how the working directory resolves in WSL2. If you use a multi-root .code-workspace, try opening a single folder directly (or the parent folder), and vice-versa.
Launch once with cursor --disable-extensions to rule out an extension stalling startup.
We’ll keep you updated on the status of the bug, and apologies for the inconvenience here
In case this helps anyone, i was able to apply a blocked fix (Execution backend unavailable) by opening a new agent window and asking it to apply the planned fix, read/write were fine in the new agent. This was on the advice of the blocked agent
So I’m using Cursor with WSL2 and
My cursor has just stopped working, and it stays frozen and unresponsive for an extensive amount of time before I can start working on it.
I’m getting below error.
Still intermittent outage — metimes, but edits/read works sosearch are still failing (Execution backend unavailable), so I can’t reliably apply changes myself yet.
I’m blocked: the execution backend isn’t available right now, so I can’t read master: via git, diff master…HEAD, or scan “/path/to/project” yet. Once tooling is back, I’ll generate the full Step 1–4 outputs exactly in your formats.
Here is a workaround until the error is fixed. I’ve created a skill that tells the agent not to run any tools, but instead instructs me to perform the task manually. Feel free to adjust it to your own taste (I’m still improving it):
