Summary
Headless browser screenshot generation can partially work in the FoxFang runtime, but the resulting screenshot file is not reliably deliverable or accessible for follow-up use.
In practice, this means an agent may be able to render a page with Chromium, but then fail to:
- save the screenshot to the expected workspace path
- verify the file exists afterward
- attach or send the generated image back to the user
This breaks the end-to-end screenshot workflow.
Problem
FoxFang currently appears to have a runtime bug or environment mismatch affecting browser screenshot output and file delivery.
During practical use with chromium-browser in headless mode against https://reply.cash, the following behavior was observed:
- Chromium could launch and render the page.
- Writing directly to the workspace path failed with a permission error.
- Writing to
/tmp/...png reported success (... bytes written to file ...).
- Immediately afterward, the file was not available for normal access or delivery.
- As a result, the agent could not send the screenshot back to the user.
This suggests the problem is not only browser launch, but the full pipeline of:
render -> write file -> verify file -> attach/deliver image
Reproduction
Example command used:
chromium-browser --headless --no-sandbox --disable-gpu --hide-scrollbars --window-size=1440,2000 --screenshot=/tmp/reply-cash-homepage.png https://reply.cash
Observed runtime behavior included:
Failed to write file /root/.foxfang/workspace/reply-cash-homepage.png: Permission denied (13)
- temporary path write appearing to succeed:
12904 bytes written to file /tmp/reply-cash-homepage.png
- immediate access failure afterward:
ls: cannot access '/tmp/reply-cash-homepage.png': No such file or directory
Expected behavior
When an agent generates a screenshot using a supported browser workflow, the runtime should allow the full workflow to succeed:
- launch browser
- render target page
- write screenshot file to a stable accessible path
- verify the file exists
- allow the image to be read/attached/sent back to the user
Actual behavior
The browser can partially execute, but the screenshot artifact is not reliably available afterward.
As a result, screenshot requests fail at the final delivery step even though rendering appears to have happened.
Why this matters
This blocks important agent workflows such as:
- taking product screenshots for users
- capturing landing pages for review
- creating visual references for marketing work
- comparing UI states across builds
- generating screenshots for bug reports and documentation
A screenshot workflow that renders but cannot deliver output is effectively broken for end users.
Possible root cause areas
Potential areas worth investigating:
- sandbox/runtime restrictions around browser file output
- path permission mismatches between workspace and browser runtime
- temporary filesystem isolation or cleanup behavior
- Snap Chromium confinement behavior
- artifact handoff between browser execution and FoxFang file tools
- attachment/delivery support for runtime-generated image files
Requested fix
Make screenshot/browser output reliably usable inside FoxFang by ensuring that agents can:
- generate screenshots to a writable path
- verify those files exist afterward
- read the files through normal tools
- attach or deliver the images back to users
Acceptance criteria
- a headless browser can generate a screenshot in FoxFang without path or permission failure
- the generated file remains accessible after command completion
- the file can be read or attached through the normal runtime toolchain
- agents can successfully send the resulting screenshot back to the user
- documentation reflects the supported screenshot path and workflow
Context
This issue comes from a real request to take a screenshot of https://reply.cash and send it to the user from within FoxFang. Rendering partially worked, but file output and delivery did not complete successfully.
Summary
Headless browser screenshot generation can partially work in the FoxFang runtime, but the resulting screenshot file is not reliably deliverable or accessible for follow-up use.
In practice, this means an agent may be able to render a page with Chromium, but then fail to:
This breaks the end-to-end screenshot workflow.
Problem
FoxFang currently appears to have a runtime bug or environment mismatch affecting browser screenshot output and file delivery.
During practical use with
chromium-browserin headless mode againsthttps://reply.cash, the following behavior was observed:/tmp/...pngreported success (... bytes written to file ...).This suggests the problem is not only browser launch, but the full pipeline of:
render -> write file -> verify file -> attach/deliver image
Reproduction
Example command used:
Observed runtime behavior included:
Expected behavior
When an agent generates a screenshot using a supported browser workflow, the runtime should allow the full workflow to succeed:
Actual behavior
The browser can partially execute, but the screenshot artifact is not reliably available afterward.
As a result, screenshot requests fail at the final delivery step even though rendering appears to have happened.
Why this matters
This blocks important agent workflows such as:
A screenshot workflow that renders but cannot deliver output is effectively broken for end users.
Possible root cause areas
Potential areas worth investigating:
Requested fix
Make screenshot/browser output reliably usable inside FoxFang by ensuring that agents can:
Acceptance criteria
Context
This issue comes from a real request to take a screenshot of
https://reply.cashand send it to the user from within FoxFang. Rendering partially worked, but file output and delivery did not complete successfully.