fix(gateway): /profile reports the profile serving the source on multiplexed gateways

[CocaKova]

On a multiplexed gateway the process-level active profile is always the multiplexer's own (usually default), so /profile answers "default" in every chat regardless of which profile actually serves it. When personas are routed per-chat (/p/<profile>/ URL prefix, per-credential adapters, or per-chat mapping as in #61689), this makes working routing look broken — /profile is the first thing an operator reaches for to check it.

When gateway.multiplex_profiles is on, this reports source.profile when stamped and resolves the displayed home under that profile's runtime scope, mirroring the scoped /reset//new banner from #59003. The source-profile lookup and scope entry are gated on multiplex_profiles — the same gating as _run_agent and _reset_notice_session_info — so with multiplexing off (the default) a stamped source is ignored and single-profile gateways are byte-for-byte unchanged.

Tests: stamped source on a multiplexed gateway reports the profile's name and profiles/<name> home; stamped source with multiplexing off is ignored (regression); unstamped source unchanged; the existing custom-root /profile test still passes.

🤖 Generated with Claude Code

[teknium1]

Thanks for targeting a real multiplexing diagnostics gap: current main's /profile still reads only the process-active profile at gateway/slash_commands.py:329-342.

Problems

• The new handler uses source.profile and enters _profile_runtime_scope without checking multiplex_profiles. Existing consumers deliberately gate this behavior: gateway/session.py:1260-1269, gateway/run.py:12200-12203, and gateway/run.py:16835-16846.
• The new stamped-source test constructs the default gateway config through _make_runner() (tests/gateway/test_status_command.py:34-40), while GatewayConfig.multiplex_profiles defaults to False (gateway/config.py:701). It does not test the intended multiplex path.

Suggested changes

• Gate the source-profile lookup and runtime scope on self.config.multiplex_profiles, preserving the existing single-profile path.
• Enable multiplexing in the stamped-source test and add a regression showing a stamped source remains ignored when multiplexing is off.

Automated hermes-sweeper review.

[CocaKova]

Both points addressed in 2e93839:

• The source-profile lookup and _profile_runtime_scope entry are now gated on self.config.multiplex_profiles (same getattr pattern as _run_agent and _reset_notice_session_info). With multiplexing off, a stamped source is ignored and /profile reports the active profile and default home, byte-identical to pre-PR behavior.
• The stamped-source test now sets multiplex_profiles = True, and a new regression (test_profile_command_ignores_stamp_when_multiplexing_off) asserts the stamp is ignored under the default config — verified it fails against the ungated handler.

[teknium1]

Merged via PR #65700 — both commits cherry-picked onto current main with your authorship preserved in git log (rebase merge). Verified before merging that the bug still reproduced: your test failed on main for URL-prefix and routing-rule stamped sources (secondary-adapter chats were partially masked by the handler scope wrap, but the stamped-source path was fully broken). Thanks!