Merge pull request #101 from abhi1693/riya/security-agent-token-regression

test(backend): xfail regression sketch for agent token lookup DoS

backend/tests/test_agent_auth_token_lookup_regression.py

@@ -0,0 +1,54 @@
+# ruff: noqa: INP001
+"""Regression test sketch for agent-token lookup complexity.
+
+Context:
+- Current implementation performs PBKDF2 verification in a loop over *all* agents
+  that have a token hash (`agent_auth._find_agent_for_token`).
+- This is O(N_agents) *and* each verify is expensive (PBKDF2 200k iterations).
+
+This test is marked xfail to document the desired behavior after a hardening
+refactor: O(1) lookup + single hash verify.
+
+Once token lookup is refactored, flip this to a normal passing test.
+"""
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+
+import pytest
+
+from app.core import agent_auth
+
+
+@pytest.mark.asyncio
+@pytest.mark.xfail(
+    reason="Known DoS risk: agent token verification is currently O(N_agents)."
+    " Refactor token scheme/lookup to O(1) and make this pass.",
+    strict=False,
+)
+async def test_agent_token_lookup_should_not_verify_more_than_once(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    # Fake a session that returns many agents with token hashes.
+    class _FakeSession:
+        async def exec(self, _stmt: object) -> list[object]:
+            agents = []
+            for i in range(50):
+                agents.append(SimpleNamespace(agent_token_hash=f"pbkdf2_sha256$1$salt{i}$digest{i}"))
+            return agents
+
+    calls = {"n": 0}
+
+    def _fake_verify(_token: str, _stored_hash: str) -> bool:
+        calls["n"] += 1
+        # Always invalid
+        return False
+
+    monkeypatch.setattr(agent_auth, "verify_agent_token", _fake_verify)
+
+    out = await agent_auth._find_agent_for_token(_FakeSession(), "invalid")  # type: ignore[arg-type]
+    assert out is None
+
+    # Desired behavior after refactor: avoid linear scan.
+    assert calls["n"] <= 1