Linux sandbox egress proxy ignores NO_PROXY when chaining to a corporate upstream proxy #5199

Describe the bug

My environment uses a corporate proxy, with HTTPS_PROXY=http://proxy.example.org:8080 and NO_PROXY=localhost,127.0.0.1,.example.org. Internal hosts such as internal.example.org are reached directly, not through the proxy.

Outside the sandbox this works. The CLI process connects directly to internal hosts, and curl returns the expected HTTP status and response.

Inside the sandbox, shell commands are pointed at the sandbox egress proxy (https_proxy=http://10.0.2.2:<port>), and NO_PROXY is not present in the env (as expected). Requests to internal hosts fail:

• curl --> OpenSSL SSL_connect: SSL_ERROR_SYSCALL (the tunnel returns 200 and then closes), or CONNECT tunnel failed, response 502
• --noproxy '*' --> could not resolve host, so direct access isn't possible either (as expected)

So the sandbox proxy chains to the corporate upstream proxy and sends internal hosts there. The upstream proxy can't route them and returns 502. The proxy also doesn't apply NO_PROXY (.example.org) to connect directly as it should.

Steps to reproduce

1. set HTTPS_PROXY to corporate proxy that can't reach internal hosts, and set NO_PROXY=.example.org
2. start copilot with sandbox enabled
3. run curl -v https://internal.example.org/ in a shell command
4. observe the 502 or SSL_ERROR_SYSCALL failure. Disable the sandbox and the same command works

Expected behavior

The sandbox egress proxy honours NO_PROXY and connects directly to matching hosts. Alternatively, it should expose a setting for hosts that bypass the upstream proxy.

Affected version

1.0.100

Steps to reproduce the behavior

No response

Expected behavior

No response

Additional context

No response