feat(swarm): self-destruct agent on decommission

Decommissioning a worker from the dashboard (or swarm controller) now
asks the agent to wipe its own install before the master forgets it.
The agent stops decky containers + every decnet-* systemd unit, then
deletes /opt/decnet*, /etc/systemd/system/decnet-*, /var/lib/decnet/*,
and /usr/local/bin/decnet*. Logs under /var/log are preserved.

The reaper runs as a detached /tmp script (start_new_session=True) so
it survives the agent process being killed. Self-destruct dispatch is
best-effort — a dead worker doesn't block master-side cleanup.

tests/swarm/test_agent_app.py

@@ -42,4 +42,49 @@ def test_deploy_rejects_malformed_body() -> None:
 
 def test_route_set() -> None:
     paths = {r.path for r in app.routes if hasattr(r, "path")}
-    assert {"/health", "/status", "/deploy", "/teardown", "/mutate"} <= paths
+    assert {"/health", "/status", "/deploy", "/teardown", "/mutate", "/self-destruct"} <= paths
+
+
+def test_self_destruct_spawns_reaper_and_returns_fast(monkeypatch, tmp_path) -> None:
+    """/self-destruct must write the reaper script and spawn it detached
+    (start_new_session=True). We intercept Popen so the test doesn't
+    actually nuke anything."""
+    from decnet.agent import executor as _exec
+
+    spawned: list[dict] = []
+
+    class _FakePopen:
+        def __init__(self, args, **kw):
+            spawned.append({"args": args, "kw": kw})
+
+    monkeypatch.setattr(_exec, "_deployer", type("X", (), {
+        "teardown": staticmethod(lambda _id: None),
+    })())
+    monkeypatch.setattr(_exec, "clear_state", lambda: None)
+
+    import subprocess as _sp
+    monkeypatch.setattr(_sp, "Popen", _FakePopen)
+
+    client = TestClient(app)
+    resp = client.post("/self-destruct")
+    assert resp.status_code == 200
+    assert resp.json()["status"] == "self_destruct_scheduled"
+    assert len(spawned) == 1
+    assert spawned[0]["kw"].get("start_new_session") is True
+    script_path = spawned[0]["args"][1]
+    assert script_path.startswith("/tmp/decnet-reaper-")
+    # Reaper content sanity check — covers the paths the operator asked for.
+    import pathlib
+    body = pathlib.Path(script_path).read_text()
+    assert "/opt/decnet*" in body
+    assert "/etc/systemd/system/decnet-" in body
+    assert "/var/lib/decnet/*" in body
+    assert "/usr/local/bin/decnet*" in body
+    # Logs must be preserved — no `rm` line should touch /var/log.
+    for line in body.splitlines():
+        stripped = line.strip()
+        if stripped.startswith("#") or not stripped:
+            continue
+        if stripped.startswith("rm "):
+            assert "/var/log" not in stripped
+    pathlib.Path(script_path).unlink(missing_ok=True)

tests/swarm/test_swarm_api.py

@@ -158,6 +158,60 @@ def test_decommission_removes_host_and_bundle(
     assert not bundle_dir.exists()
 
 
+def test_decommission_dispatches_self_destruct_to_agent(
+    client: TestClient, ca_dir: pathlib.Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Decommission must ask the worker to wipe its own install. Otherwise
+    the agent keeps running after the dashboard forgets it exists."""
+    calls: list[str] = []
+
+    class _SelfDestructAgent:
+        def __init__(self, host=None, **_):
+            self._host = host or {}
+
+        async def __aenter__(self): return self
+        async def __aexit__(self, *exc): return None
+
+        async def self_destruct(self):
+            calls.append(self._host.get("name") or "?")
+            return {"status": "self_destruct_scheduled"}
+
+    from decnet.web.router.swarm import api_decommission_host as decom_mod
+    monkeypatch.setattr(decom_mod, "AgentClient", _SelfDestructAgent)
+
+    reg = client.post(
+        "/swarm/enroll",
+        json={"name": "worker-nuke", "address": "10.0.0.8", "agent_port": 8765},
+    ).json()
+    resp = client.delete(f"/swarm/hosts/{reg['host_uuid']}")
+    assert resp.status_code == 204
+    assert calls == ["worker-nuke"]
+
+
+def test_decommission_proceeds_when_agent_unreachable(
+    client: TestClient, ca_dir: pathlib.Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """A dead worker must not block the operator from cleaning up the
+    dashboard. Self-destruct failure is logged, decommission proceeds."""
+    class _DeadAgent:
+        def __init__(self, host=None, **_): pass
+        async def __aenter__(self): return self
+        async def __aexit__(self, *exc): return None
+        async def self_destruct(self):
+            raise RuntimeError("connection refused")
+
+    from decnet.web.router.swarm import api_decommission_host as decom_mod
+    monkeypatch.setattr(decom_mod, "AgentClient", _DeadAgent)
+
+    reg = client.post(
+        "/swarm/enroll",
+        json={"name": "worker-dead", "address": "10.0.0.8", "agent_port": 8765},
+    ).json()
+    resp = client.delete(f"/swarm/hosts/{reg['host_uuid']}")
+    assert resp.status_code == 204
+    assert client.get(f"/swarm/hosts/{reg['host_uuid']}").status_code == 404
+
+
 # ---------------------------------------------------------------- /deploy