refactor(integrations): shared agent plugin runtimes and native adapters (#7203)

integrations/claude-code-plugin/tests/conftest.py

@@ -1,7 +1,20 @@
-"""Keep the test suite from sending usage telemetry to the live PostHog project."""
+"""Configure harness identity and disable live telemetry for the test suite."""
 
 from __future__ import annotations
 
 import os
+import sys
+from pathlib import Path
 
 os.environ["MEM0_TELEMETRY"] = "false"
+
+HOST_ROOT = Path(__file__).resolve().parents[1]
+_core = HOST_ROOT / "core"
+sys.path.insert(0, str(_core))
+
+import hook_runner  # noqa: E402,F401
+from memory_core import configure_harness  # noqa: E402
+import telemetry  # noqa: E402
+
+configure_harness("claude-code", data_dir_name="claude-code-plugin", source_tag="claude_code_plugin")
+telemetry.init(harness="claude-code", source_tag="CLAUDE_CODE_PLUGIN")

integrations/claude-code-plugin/tests/integration/test_live_scoping.py

@@ -13,10 +13,11 @@ from pathlib import Path
 import pytest
 
 PLUGIN_ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(PLUGIN_ROOT / "core"))
-sys.path.insert(0, str(PLUGIN_ROOT / "adapters" / "claude"))
+_core_dir = PLUGIN_ROOT / "core"
+sys.path.insert(0, str(_core_dir))
+sys.path.insert(0, str(PLUGIN_ROOT))
 
-import hook  # noqa: E402
+import hook_runner  # noqa: E402
 import memory_core  # noqa: E402
 
 pytestmark = pytest.mark.skipif(
@@ -177,15 +178,17 @@ def test_shared_project_memory_reaches_a_teammate_who_never_wrote(ns):
 def test_repo_scope_spans_every_subdirectory(ns):
     found = ns.search("erin", ns.root, "how many times are Stripe webhooks retried", scope="repo")
     assert "stripe" in _text(found)
-    app_ids = {m.get("app_id") for m in found}
-    assert any(app.endswith("/services/billing") for app in app_ids)
+    assert any("services/billing" in (memory.get("metadata") or {}).get("dirs", []) for memory in found)
 
 
 def test_dir_scope_narrows_shared_memory_to_the_directory(ns):
     billing = ns.root / "services" / "billing"
     found = ns.search("erin", billing, "how do I run the tests here", scope="dir")
     assert "billing-test" in _text(found)
-    assert {m.get("app_id") for m in found if m.get("agent_id")} == {memory_core.directory_app_id(memory_core.resolve_repo(str(billing)))}
+    shared = [memory for memory in found if memory.get("agent_id")]
+    assert shared
+    assert {memory.get("app_id") for memory in shared} == {memory_core.resolve_repo(str(billing)).app_id}
+    assert all("services/billing" in (memory.get("metadata") or {}).get("dirs", []) for memory in shared)
 
     web = ns.root / "apps" / "web"
     elsewhere = ns.search("erin", web, "Stripe webhook retries exponential backoff", scope="dir", tries=1)
@@ -232,7 +235,7 @@ def test_a_pending_packet_is_recovered_and_delivered_by_the_worker(ns):
     stale.write_text(json.dumps({"hook_input": {"session_id": sid, "cwd": str(ns.root)}, "reason": "session-end"}))
     os.utime(stale, (time.time() - 3600, time.time() - 3600))
 
-    assert hook.recover_pending_handoffs() == 1
+    assert hook_runner.recover_pending_handoffs() == 1
     deadline = time.time() + 240
     while time.time() < deadline and list(pending.iterdir()):
         time.sleep(3)

integrations/claude-code-plugin/tests/test_claude_build.py

@@ -0,0 +1,35 @@
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+HOST = Path(__file__).resolve().parents[1]
+CORE_ROOT = HOST.parent / "agent-plugin-core"
+sys.path.insert(0, str(CORE_ROOT))
+
+from build.build import SHARED_SKILLS, build, render_template  # noqa: E402
+
+
+def test_native_claude_bundle_preserves_working_contract(tmp_path: Path) -> None:
+    root = build("claude-code", "native", tmp_path / "claude-code")
+
+    assert (root / ".claude-plugin" / "plugin.json").read_bytes() == (
+        HOST / ".claude-plugin" / "plugin.json"
+    ).read_bytes()
+    assert (root / ".mcp.json").read_bytes() == (HOST / ".mcp.json").read_bytes()
+    assert (root / "hooks" / "hooks.json").read_bytes() == (HOST / "hooks" / "hooks.json").read_bytes()
+    assert (root / "agents" / "sidekick.md").read_bytes() == (HOST / "agents" / "sidekick.md").read_bytes()
+    values = {
+        "PLUGIN_ROOT": "${CLAUDE_PLUGIN_ROOT}",
+        "PLUGIN_DATA": "${PLUGIN_DATA}",
+        "PLUGIN_DATA_ARG": '--plugin-data-dir "${CLAUDE_PLUGIN_DATA}"',
+        "COMMAND_PREFIX": "mem0",
+        "HARNESS_ID": "claude-code",
+        "HARNESS_NAME": "Claude Code",
+    }
+    for skill in SHARED_SKILLS.glob("*/SKILL.md.tmpl"):
+        rendered = render_template(skill.read_text(encoding="utf-8"), values)
+        assert (root / "skills" / skill.parent.name / "SKILL.md").read_text(encoding="utf-8") == rendered
+    assert (root / "adapters" / "claude" / "hook.py").is_file()
+    assert (root / "core" / "mcp_server.py").is_file()
+    assert not any(path.is_symlink() for path in root.rglob("*"))

integrations/claude-code-plugin/tests/test_telemetry.py

@@ -7,9 +7,8 @@ from unittest.mock import patch
 
 import pytest
 
-
-PLUGIN_ROOT = Path(__file__).resolve().parents[1]
-CORE = PLUGIN_ROOT / "core"
+HOST_ROOT = Path(__file__).resolve().parents[1]
+CORE = HOST_ROOT / "core"
 sys.path.insert(0, str(CORE))
 
 import memory_core  # noqa: E402
@@ -77,6 +76,46 @@ def test_record_hashes_identifiers_and_keeps_no_content(isolated_env):
     assert "session-abcdef" not in serialized
 
 
+def test_record_rejects_sensitive_properties_at_the_shared_boundary(isolated_env):
+    secret = "sk-eval-12345678901234567890"
+    telemetry.record(
+        "search",
+        prompt=f"remember {secret}",
+        query=secret,
+        api_key=secret,
+        user_id="private-user",
+        note=f"failure contained {secret}",
+        memory_count=2,
+    )
+
+    (event,) = spool_lines()
+    assert event["properties"]["memory_count"] == 2
+    serialized = json.dumps(event)
+    assert secret not in serialized
+    assert "private-user" not in serialized
+    assert not {"prompt", "query", "api_key", "user_id"} & event["properties"].keys()
+
+
+@pytest.mark.parametrize("key", ["password", "token", "secret", "authorization"])
+def test_record_removes_sensitive_keys_from_nested_lists(isolated_env, key):
+    secret = "sk-eval-12345678901234567890"
+    telemetry.record(
+        "search",
+        details=[
+            {key: "plain-value", "count": 2},
+            {"nested": {key.upper(): "plain-value", "ok": True}},
+            [f"failure contained {secret}"],
+        ],
+    )
+
+    (event,) = spool_lines()
+    assert event["properties"]["details"] == [
+        {"count": 2},
+        {"nested": {"ok": True}},
+        ["failure contained [REDACTED]"],
+    ]
+
+
 def test_record_stops_appending_past_the_spool_cap(isolated_env):
     spool = memory_core.data_dir() / "telemetry.jsonl"
     spool.parent.mkdir(parents=True, exist_ok=True)