test fixtures vs live constellation

This one is simple. A clone test hardcoded agent7. A real agent7 was already alive. The live constellation ate the fixture (git show 72c0cd9, src/crates/netsky-cli/src/cmd/clone.rs:1125, src/crates/netsky-cli/src/cmd/clone.rs:1175).

flowchart LR
    A[test fixture: agent7] --> C[shared tmux server]
    B[real clone: agent7] --> C
    C --> D[test loses]

bin/check went red — not because the clone-brief code was broken, but because the test tried to dispatch a clone and hit a real pane. The code path did the right thing and left the brief on the bus when no pane settled and no ack arrived (src/crates/netsky-cli/src/cmd/clone.rs:190).

In a unit test, agent7 looks like a harmless string. In a live system, it’s a resource. The commit message for 72c0cd9 spells it out: the dispatch is-up check found a real agent7, so the test failed with agent 'agent7' is already up while real work was in flight (git show 72c0cd9).

The fix was to stop borrowing names from the live pool. 72c0cd9 moves the fixture to agent8888 and updates the roundtrip tests to match (src/crates/netsky-cli/src/cmd/clone.rs:1125-1129, src/crates/netsky-cli/src/cmd/clone.rs:1175-1185).

-                requested_agent: Some(7),
+                requested_agent: Some(8888),
...
-        assert_eq!(result.agent, "agent7");
+        assert_eq!(result.agent, "agent8888");
...
-        run_roundtrip_with_mock_clone(unique_test_session_name("agent7"))
+        run_roundtrip_with_mock_clone(unique_test_session_name("agent8888"))

agent8888 is unremarkable on purpose: far above any normal constellation size, obvious in logs, hard to confuse with a real clone. The inline comment says so in the code (src/crates/netsky-cli/src/cmd/clone.rs:1125-1129).

Small bug, but a live-system one. The test suite reached outside its fixture world, touched a real pane, and lost — the kind of failure that only shows up when the system is busy enough for production state to fight back.