chore: merge origin/main into mintlify/af82bda4 and resolve conflicts

This commit is contained in:
mintlify[bot]
2026-09-16 11:46:52 +00:00
committed by GitHub
parent b5ae772eae
commit 08d9636491
693 changed files with 55668 additions and 22676 deletions
@@ -0,0 +1,20 @@
"""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")
@@ -0,0 +1,245 @@
"""Live scoping tests against the Mem0 Platform: run with MEM0_API_KEY set, skipped otherwise."""
from __future__ import annotations
import json
import os
import subprocess
import sys
import time
import uuid
from pathlib import Path
import pytest
PLUGIN_ROOT = Path(__file__).resolve().parents[2]
_core_dir = PLUGIN_ROOT / "core"
sys.path.insert(0, str(_core_dir))
sys.path.insert(0, str(PLUGIN_ROOT))
import hook_runner # noqa: E402
import memory_core # noqa: E402
pytestmark = pytest.mark.skipif(
not os.environ.get("MEM0_API_KEY"), reason="MEM0_API_KEY is required for live scoping tests"
)
GIT_ENV = {
**os.environ,
"GIT_AUTHOR_NAME": "t",
"GIT_AUTHOR_EMAIL": "t@t",
"GIT_COMMITTER_NAME": "t",
"GIT_COMMITTER_EMAIL": "t@t",
}
def _git_repo(path: Path, remote: str) -> None:
path.mkdir(parents=True)
for args in (
["init", "-q"],
["remote", "add", "origin", remote],
["commit", "-q", "--allow-empty", "-m", "init"],
):
subprocess.run(["git", "-C", str(path), *args], check=True, capture_output=True, env=GIT_ENV)
def _bash(command: str, failed: bool, preview: str) -> dict:
return {
"tool": "Bash",
"command": command,
"command_kind": "test",
"failed": failed,
"result_preview": preview,
}
class Namespace:
"""One fresh, isolated set of users and repositories for a test run."""
def __init__(self, tmp: Path):
self.tag = uuid.uuid4().hex[:8]
self.users = {name: f"live-{self.tag}-{name}" for name in ("alice", "bob", "carol", "dave", "erin")}
self.root = tmp / "monorepo"
_git_repo(self.root, f"https://github.com/live-{self.tag}/monorepo.git")
(self.root / "services" / "billing").mkdir(parents=True)
(self.root / "apps" / "web").mkdir(parents=True)
self.notes_a = tmp / "carol" / "notes"
self.notes_b = tmp / "dave" / "notes"
self.notes_a.mkdir(parents=True)
self.notes_b.mkdir(parents=True)
os.environ["MEM0_CODE_DATA_DIR"] = str(tmp / "data")
os.environ["MEM0_CODE_TELEMETRY"] = "false"
os.environ["MEM0_CODE_EXTRACTION_WAIT_SECONDS"] = "180"
self.store = memory_core.EvidenceStore()
self.reads = 0
def as_user(self, name: str) -> str:
os.environ["MEM0_CODE_USER_ID"] = self.users[name]
return self.users[name]
def session(self, user: str, cwd: Path, sid: str, prompt: str, tools: list[dict], answer: str, reason="session-end"):
self.as_user(user)
sid = f"{self.tag}-{sid}"
ctx = self.store.repo_for_session(sid, str(cwd))
self.store.record_event(ctx, sid, "user_prompt", {"text": prompt})
for tool in tools:
self.store.record_event(ctx, sid, "tool_result", tool)
self.store.record_event(ctx, sid, "assistant_stop", {"text": answer})
result = memory_core.flush_session(self.store, {"session_id": sid, "cwd": str(cwd)}, reason)
assert result.get("status") == "semantic-succeeded", result
return sid
def search(self, user: str, cwd: Path, query: str, *, tries: int = 6, top_k: int = 20, **kwargs) -> list[dict]:
self.as_user(user)
ctx = memory_core.resolve_repo(str(cwd))
memories: list[dict] = []
for attempt in range(tries):
self.reads += 1
memories = memory_core.search_memories(
self.store, ctx, f"{self.tag}-read-{self.reads}", query, top_k=top_k, operation="live-test", timeout=30, **kwargs
).memories
if memories or attempt == tries - 1:
return memories
time.sleep(5)
return memories
def cleanup(self):
for user, cwd in (("alice", self.root), ("bob", self.root), ("erin", self.root), ("carol", self.notes_a), ("dave", self.notes_b)):
self.as_user(user)
memory_core.forget_remote_repo(memory_core.resolve_repo(str(cwd)), include_project_memory=True)
self.store.close()
def _text(memories: list[dict]) -> str:
return " ".join(str(m.get("memory", "")) for m in memories).lower()
@pytest.fixture(scope="module")
def ns(tmp_path_factory):
namespace = Namespace(tmp_path_factory.mktemp("live"))
root, billing, web = namespace.root, namespace.root / "services" / "billing", namespace.root / "apps" / "web"
namespace.session(
"alice", root, "root-alice",
"Remember that I personally prefer uv over pip. Also document that invoices are rounded half-up in api/invoices.py.",
[{"tool": "Edit", "path": "README.md"}, _bash("pytest", False, "12 passed")],
"README now documents that invoices round half-up in api/invoices.py. Noted that you prefer uv over pip.",
)
namespace.session(
"bob", billing, "billing-bob",
"The billing worker must retry Stripe webhooks five times with exponential backoff. Run the billing tests.",
[_bash("npm test", True, "npm ERR! missing script: test"), _bash("make billing-test", False, "34 passed")],
"Documented: the billing worker retries Stripe webhooks five times with exponential backoff. `npm test` does not exist here; `make billing-test` runs the billing suite.",
)
namespace.session(
"bob", web, "web-bob",
"The web app is built with Vite. Start it with pnpm --filter web dev.",
[_bash("pnpm --filter web dev", False, "VITE ready in 300ms")],
"Confirmed: the web app uses Vite and starts with `pnpm --filter web dev`.",
)
namespace.session(
"carol", namespace.notes_a, "notes-carol",
"Private notes folder. My journal password hint lives in hints.txt. I like vim keybindings.",
[{"tool": "Edit", "path": "hints.txt"}],
"Added the journal password hint to hints.txt. Noted that you like vim keybindings.",
)
namespace.session(
"dave", namespace.notes_b, "notes-dave",
"This notes folder holds my grocery list in groceries.md.",
[{"tool": "Edit", "path": "groceries.md"}],
"Saved the grocery list to groceries.md.",
)
namespace.handoff_session = namespace.session(
"alice", root, "handoff-old",
"We decided to migrate the ledger table to bigint ids. Migration 0042 is written but test_ledger_precision still fails.",
[_bash("pytest tests/test_ledger.py", True, "FAILED test_ledger_precision: Decimal rounding mismatch")],
"Migration 0042 moves ledger ids to bigint. test_ledger_precision still fails with a Decimal rounding mismatch; that is the next thing to fix.",
reason="pre-compact",
)
yield namespace
namespace.cleanup()
def test_personal_preferences_stay_with_their_owner(ns):
mine = ns.search("alice", ns.root, "which package manager do I prefer", scope="mine")
assert "uv" in _text(mine)
assert {m.get("user_id") for m in mine} == {ns.users["alice"]}
teammate = ns.search("bob", ns.root, "which package manager do I prefer, uv or pip", tries=1)
assert not any(m.get("user_id") == ns.users["alice"] for m in teammate)
assert "uv" not in _text(teammate)
def test_shared_project_memory_reaches_a_teammate_who_never_wrote(ns):
found = ns.search("erin", ns.root, "how are invoices rounded")
assert "half" in _text(found)
assert all(m.get("user_id") is None and m.get("agent_id") for m in found)
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)
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)
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)
assert "stripe" not in _text(elsewhere)
def test_project_memory_never_carries_a_user_id(ns):
found = ns.search("erin", ns.root, "invoices rounding billing webhooks vite dev server")
assert found
for memory in found:
assert memory.get("user_id") is None
assert memory.get("agent_id") == memory_core.resolve_repo(str(ns.root)).project_id
def test_same_named_plain_folders_at_different_paths_do_not_share(ns):
carol = ns.search("carol", ns.notes_a, "where is my journal password hint")
assert "hint" in _text(carol)
dave = ns.search("dave", ns.notes_b, "where is the journal password hint", tries=1)
assert "hint" not in _text(dave)
assert not any((m.get("metadata") or {}).get("author") == ns.users["carol"] for m in dave)
def test_run_id_recovers_one_session_after_compaction(ns):
found = ns.search("alice", ns.root, "what was I working on and what still fails", run_id=ns.handoff_session)
assert "ledger" in _text(found) or "0042" in _text(found)
assert {m.get("run_id") for m in found} == {ns.handoff_session}
other = ns.search("alice", ns.root, "invoices rounding half-up", run_id=ns.handoff_session, tries=1)
assert all(m.get("run_id") == ns.handoff_session for m in other)
def test_a_pending_packet_is_recovered_and_delivered_by_the_worker(ns):
ns.as_user("alice")
sid = f"{ns.tag}-recovered"
ctx = ns.store.repo_for_session(sid, str(ns.root))
ns.store.record_event(ctx, sid, "user_prompt", {"text": "Note that nightly builds are published from the release-bot machine at 02:00 UTC."})
ns.store.record_event(ctx, sid, "tool_result", {"tool": "Edit", "path": "docs/releases.md"})
ns.store.record_event(ctx, sid, "assistant_stop", {"text": "Documented that nightly builds are published from the release-bot machine at 02:00 UTC."})
pending = memory_core.data_dir() / "pending"
pending.mkdir(parents=True, exist_ok=True)
stale = pending / "stale-run.running"
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_runner.recover_pending_handoffs() == 1
deadline = time.time() + 240
while time.time() < deadline and list(pending.iterdir()):
time.sleep(3)
assert not list(pending.iterdir()), "worker left its packet behind"
found = ns.search("erin", ns.root, "when and where are nightly builds published", run_id=sid)
assert "nightly" in _text(found) or "02:00" in _text(found)
@@ -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("*"))
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,275 @@
from __future__ import annotations
import json
import sys
from pathlib import Path
from unittest.mock import patch
import pytest
HOST_ROOT = Path(__file__).resolve().parents[1]
CORE = HOST_ROOT / "core"
sys.path.insert(0, str(CORE))
import memory_core # noqa: E402
import telemetry # noqa: E402
@pytest.fixture
def isolated_env(tmp_path, monkeypatch):
monkeypatch.setenv("MEM0_CODE_DATA_DIR", str(tmp_path / "data"))
monkeypatch.setenv("MEM0_TELEMETRY", "true")
monkeypatch.delenv("MEM0_API_KEY", raising=False)
monkeypatch.delenv("CLAUDE_PLUGIN_OPTION_API_KEY", raising=False)
monkeypatch.delenv("CLAUDE_PLUGIN_OPTION_MEM0_API_KEY", raising=False)
monkeypatch.delenv("MEM0_API_URL", raising=False)
return tmp_path
def repo() -> memory_core.RepoContext:
return memory_core.RepoContext(
cwd="/tmp/repo",
root="/tmp/repo",
identity="https://github.com/example/secret-repo",
app_id="code-example",
branch="main",
head_sha="abc123",
)
def spool_lines() -> list[dict]:
path = memory_core.data_dir() / "telemetry.jsonl"
if not path.exists():
return []
return [json.loads(line) for line in path.read_text().splitlines()]
def test_opt_out_writes_nothing(isolated_env, monkeypatch):
for value in ("false", "0", "no", "OFF"):
monkeypatch.setenv("MEM0_TELEMETRY", value)
telemetry.record("search", repo=repo(), session_id="s-1")
assert not telemetry.is_enabled()
assert spool_lines() == []
def test_record_hashes_identifiers_and_keeps_no_content(isolated_env):
telemetry.record(
"search",
repo=repo(),
session_id="session-abcdef",
trigger="first-prompt-search",
matched_count=3,
dropped=None,
)
(event,) = spool_lines()
assert event["event"] == "code.search"
assert event["timestamp"]
properties = event["properties"]
assert properties["harness"] == "claude-code"
assert properties["plugin_version"] == memory_core.PLUGIN_VERSION
assert properties["matched_count"] == 3
assert "dropped" not in properties
assert len(properties["repo_hash"]) == 16
assert len(properties["session_hash"]) == 16
serialized = json.dumps(event)
assert "secret-repo" not in serialized
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)
spool.write_text("x" * (telemetry.SPOOL_LIMIT_BYTES + 1))
telemetry.record("search")
assert spool.read_text() == "x" * (telemetry.SPOOL_LIMIT_BYTES + 1)
def test_record_never_raises_on_a_broken_spool(isolated_env, monkeypatch):
monkeypatch.setattr(telemetry, "_spool_path", lambda: Path("/does/not/exist/x"))
telemetry.record("search")
def test_error_kind_stays_coarse_and_content_free():
assert telemetry.error_kind("HTTP 429 too many requests") == "rate-limited"
assert telemetry.error_kind("HTTP 401 for /v1/memories/") == "auth"
assert telemetry.error_kind("HTTP 503 upstream") == "server-error"
assert telemetry.error_kind(TimeoutError("timed out")) == "timeout"
assert telemetry.error_kind(ValueError("token sk-abcdef leaked")) == "ValueError"
def test_flush_posts_one_batch_and_clears_the_spool(isolated_env):
telemetry.record("session_start")
telemetry.record("search", matched_count=1)
posted = []
with patch.object(telemetry, "_post", lambda payload, url: posted.append((payload, url)) or True):
assert telemetry.flush() == 2
(payload, url) = posted[0]
assert url == telemetry.POSTHOG_BATCH_URL
assert payload["api_key"] == telemetry.POSTHOG_API_KEY
assert [event["event"] for event in payload["batch"]] == [
"code.session_start",
"code.search",
]
first = payload["batch"][0]
assert first["distinct_id"].startswith("code-anon-")
assert first["properties"]["source"] == "CLAUDE_CODE_PLUGIN"
assert first["properties"]["$process_person_profile"] is False
assert not (memory_core.data_dir() / "telemetry.jsonl").exists()
assert not list(memory_core.data_dir().glob("telemetry-*.sending"))
def test_flush_chunks_batches(isolated_env):
for index in range(telemetry.BATCH_SIZE + 5):
telemetry.record("search", index=index)
sizes = []
with patch.object(
telemetry, "_post", lambda payload, url: sizes.append(len(payload["batch"])) or True
):
assert telemetry.flush() == telemetry.BATCH_SIZE + 5
assert sizes == [telemetry.BATCH_SIZE, 5]
def test_a_failed_post_keeps_the_events_for_the_next_run(isolated_env):
telemetry.record("search")
with patch.object(telemetry, "_post", lambda payload, url: False):
assert telemetry.flush() == 0
claims = list(memory_core.data_dir().glob("telemetry-*.sending"))
assert len(claims) == 1
assert json.loads(claims[0].read_text().splitlines()[0])["event"] == "code.search"
def test_a_claimed_spool_is_not_sent_twice(isolated_env):
telemetry.record("search")
first = telemetry._claim_spool()
assert first is not None
assert telemetry._claim_spool() is None
with patch.object(telemetry, "_post", lambda payload, url: True):
assert telemetry.flush() == 0
def test_a_stale_claim_is_reclaimed(isolated_env, monkeypatch):
telemetry.record("search")
orphan = telemetry._claim_spool()
assert orphan is not None
monkeypatch.setattr(
telemetry.time, "time", lambda: orphan.stat().st_mtime + telemetry.CLAIM_STALE_SECONDS + 1
)
with patch.object(telemetry, "_post", lambda payload, url: True):
assert telemetry.flush() == 1
def test_an_expired_claim_is_dropped(isolated_env, monkeypatch):
telemetry.record("search")
orphan = telemetry._claim_spool()
assert orphan is not None
monkeypatch.setattr(
telemetry.time, "time", lambda: orphan.stat().st_mtime + telemetry.CLAIM_EXPIRY_SECONDS + 1
)
assert telemetry._claim_spool() is None
assert not list(memory_core.data_dir().glob("telemetry-*.sending"))
def test_the_email_replaces_the_anonymous_id_once_and_is_aliased(isolated_env, monkeypatch):
monkeypatch.setenv("MEM0_API_KEY", "test-key")
anonymous = telemetry.anonymous_id()
telemetry.record("search")
posted = []
with (
patch.object(telemetry, "_resolve_email", lambda key: "dev@example.com"),
patch.object(telemetry, "_post", lambda payload, url: posted.append(payload) or True),
):
assert telemetry.flush() == 1
identify, batch = posted
assert identify["event"] == "$identify"
assert identify["distinct_id"] == "dev@example.com"
assert identify["properties"]["$anon_distinct_id"] == anonymous
assert batch["batch"][0]["distinct_id"] == "dev@example.com"
telemetry.record("search")
posted.clear()
with (
patch.object(telemetry, "_resolve_email", lambda key: pytest.fail("re-resolved")),
patch.object(telemetry, "_post", lambda payload, url: posted.append(payload) or True),
):
assert telemetry.flush() == 1
assert [payload.get("event") for payload in posted] == [None]
def test_an_unresolvable_key_falls_back_to_the_anonymous_id(isolated_env, monkeypatch):
monkeypatch.setenv("MEM0_API_KEY", "test-key")
telemetry.record("search")
with (
patch.object(telemetry, "_resolve_email", lambda key: ""),
patch.object(telemetry, "_post", lambda payload, url: True),
):
assert telemetry.flush() == 1
assert telemetry.resolve_distinct_id()[0].startswith("code-anon-")
def test_is_first_run_flips_after_the_first_identity_write(isolated_env):
assert telemetry.is_first_run()
telemetry.anonymous_id()
assert not telemetry.is_first_run()
def test_spawn_flush_does_nothing_without_a_spool(isolated_env):
with patch.object(telemetry.subprocess, "Popen") as popen:
assert telemetry.spawn_flush() is False
popen.assert_not_called()
telemetry.record("search")
with patch.object(telemetry.subprocess, "Popen") as popen:
assert telemetry.spawn_flush() is True
popen.assert_called_once()