Devin CLI bypasses HTTP(S)_PROXY for agent backend connection and hangs behind corporate proxy

[woraphonedge]

Summary

Devin CLI hangs during agent initialization on a Linux server behind an authenticated corporate HTTP proxy.

The proxy environment variables are configured correctly and Devin uses the proxy for some connections, but it also opens a direct TCP connection to a public AWS endpoint on port 443.

Direct internet access is blocked in this environment, so that connection remains in SYN-SENT and Devin never finishes initializing.

It shows

>devin -p "reply with hello"
Welcome to Devin CLI!
✓ Logged in as xxxx@gmail.com.

and that's it.

Environment

• Devin CLI: 3000.10.21 (611c1cba)
• OS: Linux
• Connection: SSH
• Corporate proxy: authenticated HTTP proxy
• Direct outbound internet access: blocked
• Proxy variables configured:
  • HTTP_PROXY
  • HTTPS_PROXY
  • http_proxy
  • https_proxy
• NODE_USE_ENV_PROXY=1

devin doctor passes:

pass  custom subagent profiles  none configured

1 check(s): 1 passed, 0 warning(s), 0 failure(s)

Behavior

Running:

devin

starts the CLI, but it remains stuck during initialization.

While Devin is running:

ss -tpn | grep devin

shows both proxied and direct connections:

ESTAB    ... -> 10.99.25.44:8080
ESTAB    ... -> 10.99.25.41:8080
SYN-SENT ... -> 44.233.100.53:443

The 10.99.25.x:8080 connections are the corporate proxy.

The direct connection to 44.233.100.53:443 never establishes.

The public destination resolves to AWS:

44.233.100.53 ec2-44-233-100-53.us-west-2.compute.amazonaws.com

I have also seen the direct destination change between public AWS IPs, suggesting this is a load-balanced backend.

Reproduction / network verification

The same destination works when explicitly routed through the configured proxy:

curl --proxy "$https_proxy" --connect-timeout 10 https://44.233.100.53/

Result:

Connected to proxy...
CONNECT tunnel established
SSL certificate verify ok
HTTP/2 404

So connectivity through the corporate proxy is working.

However, bypassing the proxy reproduces the same behavior seen from Devin:

curl --noproxy '*' -v --connect-timeout 10 https://44.233.100.53/

Result:

Trying 44.233.100.53:443...
ipv4 connect timeout after 10000ms
curl: (28) Failed to connect ... Timeout was reached

Expected behavior

All external Devin backend connections should honor the configured HTTP_PROXY / HTTPS_PROXY settings.

In particular, the backend connection currently attempting:

<server> -> public AWS IP:443

should instead go through:

<server> -> corporate proxy:8080 -> Devin backend

Actual behavior

Some Devin connections correctly use the proxy, but at least one connection used during agent initialization bypasses it and attempts direct internet access.

Because direct outbound traffic is blocked, the CLI hangs.

Additional notes

• Authentication succeeds.
• devin doctor reports no connectivity problems.
• HTTPS through the proxy works.
• HTTP/2 through the proxy works.
• TLS certificate verification through the corporate proxy works.
• This appears to affect a specific networking/backend path rather than proxy configuration in general.

It would be helpful to know whether there is an explicit Devin proxy configuration for this backend transport, or whether this is a bug where one internal client does not honor the standard proxy environment variables.