Repair Half-Closed Socket Bugs

The problem

TCP proxies, tunnels, and stream adapters often treat a read EOF as a signal to kill the entire connection. This leads to truncated responses, hung writers, and race conditions when one peer finishes sending but still needs to receive data.

What it does

• Identifies incorrect socket shutdown logic in proxies, relays, and protocol bridges.
• Implements directional CloseWrite/FIN propagation to ensure data drains before cleanup.
• Refactors read loops to decouple stream completion from socket destruction.
• Builds deterministic loopback tests to verify specific FIN/EOF ordering and buffer boundaries.
• Adds observability for peer EOF, local FIN, and truncation events.

Frameworks & tools

Works with low-level networking code in Node.js (allowHalfOpen), Go (TCPConn), C/C++, and Rust. Compatible with TCP, TLS wrappers, and custom protocol tunnels.

Why this beats prompting it yourself

General LLMs often assume EOF means "the connection is over," which is fatal for many protocols. This skill enforces a strict directional socket register and synchronization contract to prevent data loss in half-closed states.

Use cases

• Fixing tunnels where the client connection drops before the server finishes responding.
• Repairing proxy relays that truncate buffered data after a peer sends a FIN.
• Resolving deadlocks in request-response flows over raw TCP sockets.
• Migrating raw TCP close semantics to TLS close_notify without breaking stream logic.

Known limitations

Cannot verify middlebox or ISP-level packet interference. Requires repository access to local socket fixtures for validation.