Summary

Gateway /resume was restoring the titled parent session after compression instead of the live continuation child that actually holds the transcript.

This aligns gateway resume behavior with the CLI by resolving the resumed session ID through SessionDB.resolve_resume_session_id() before switching sessions.

Problem

Context compression ends the current session and creates a child continuation linked via parent_session_id.

The gateway /resume flow already resolved a title to a session ID, but it did not follow the compression continuation chain afterward. That meant:

• /resume <title> could switch to the compressed parent session
• the restored transcript could be empty or stale
• gateway behavior diverged from the CLI, which already follows the continuation descendant

Fix

• Update GatewayRunner._handle_resume_command() in gateway/run.py to pass the resolved target session through resolve_resume_session_id()
• Keep the change best-effort with a debug log fallback if continuation resolution fails
• Add a regression test in tests/gateway/test_resume_command.py that creates a compressed parent + child continuation and verifies /resume switches to the child and counts its messages

How to test

uv run --extra dev pytest tests/gateway/test_resume_command.py

Result:

20 workers [10 items]
..........                                                               [100%]
============================= 10 passed in 6.24s ==============================

Manual check: in a gateway session, trigger /compress to create a continuation child, then run /resume <title> — the gateway now reopens the live child transcript instead of the compressed parent.

No platform-specific code paths were touched (no file I/O, process management, or shell command changes), so behavior should be identical on Linux and macOS.

Files changed

• gateway/run.py — route resolved session ID through resolve_resume_session_id()
• tests/gateway/test_resume_command.py — regression test for compression-continuation resume