diff --git a/AGENTS.md b/AGENTS.md
index 7e1582761b..e66437159b 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -113,6 +113,44 @@ repos/{owner}/{repo}/milestones --jq '.[].title'`, and pick the best match.
   end on clarifications unless truly blocked. Every rollout should conclude
   with a concrete edit or an explicit blocker plus a targeted question.
 
+## Test authorship markers
+
+Use pytest markers to make the provenance of newly added unit tests explicit.
+Keep this provenance system minimal: choose from only these three markers.
+Place the marker immediately above each test function. A class-level marker is
+acceptable only when every test method in the class has the same provenance.
+Do not use module-level `pytestmark` for authorship provenance; it is too easy
+to miss in large files and makes later per-test provenance changes ambiguous.
+
+- `@pytest.mark.agent_authored(model="<model>")`: the test was authored by an
+  agent and has not yet been materially reviewed or rewritten by a human.
+  Agents must add this marker when generating new unit tests, for example:
+
+  ```python
+  import pytest
+
+  @pytest.mark.agent_authored(model="gpt-5.5")
+  def test_something():
+      ...
+  ```
+
+- `@pytest.mark.human_reviewed`: a human has materially reviewed or rewritten
+  an agent-authored test. Prefer replacing
+  `@pytest.mark.agent_authored(model="<model>")` with this marker instead of
+  keeping both.
+- `@pytest.mark.human_authored`: the test was authored by a human, or
+  rewritten enough that the authorship is primarily human.
+
+Use at most one authorship marker per test. Treat missing markers as legacy or
+unknown provenance, not as implicit `@pytest.mark.human_authored`. Because
+these are pytest markers, tests can be selected with `pytest -m`, for example
+`pytest -m agent_authored`.
+
+When an agent notices a human adding a new test or materially modifying an
+existing test, suggest adding `@pytest.mark.human_authored` or replacing
+`@pytest.mark.agent_authored(model="<model>")` with
+`@pytest.mark.human_reviewed` as appropriate.
+
 
 # Editing constraints
 
diff --git a/cuda_bindings/pyproject.toml b/cuda_bindings/pyproject.toml
index c3f1529461..7b35e9c140 100644
--- a/cuda_bindings/pyproject.toml
+++ b/cuda_bindings/pyproject.toml
@@ -92,6 +92,13 @@ required_plugins = "pytest-benchmark"
 addopts = "--benchmark-disable --showlocals"
 norecursedirs = ["tests/cython", "examples"]
 xfail_strict = true
+# Keep this authorship marker registry in sync across all pytest config roots.
+# Search for "agent_authored(model)" before editing.
+markers = [
+    "agent_authored(model): agent-authored test not yet materially human-reviewed",
+    "human_reviewed: agent-authored test materially reviewed or rewritten by a human",
+    "human_authored: test authored primarily by a human",
+]
 
 [tool.setuptools_scm]
 root = ".."
diff --git a/cuda_core/pytest.ini b/cuda_core/pytest.ini
index 41bf0d9c4f..c3d243387f 100644
--- a/cuda_core/pytest.ini
+++ b/cuda_core/pytest.ini
@@ -5,3 +5,9 @@
 [pytest]
 addopts = --showlocals
 norecursedirs = cython
+markers =
+    # Keep this authorship marker registry in sync across all pytest config roots.
+    # Search for "agent_authored(model)" before editing.
+    agent_authored(model): agent-authored test not yet materially human-reviewed
+    human_reviewed: agent-authored test materially reviewed or rewritten by a human
+    human_authored: test authored primarily by a human
diff --git a/cuda_pathfinder/pyproject.toml b/cuda_pathfinder/pyproject.toml
index c457438274..9d9f025085 100644
--- a/cuda_pathfinder/pyproject.toml
+++ b/cuda_pathfinder/pyproject.toml
@@ -104,6 +104,13 @@ git_describe_command = [ "git", "describe", "--dirty", "--tags", "--long", "--ma
 [tool.pytest.ini_options]
 addopts = "--showlocals"
 thread_unsafe_fixtures = ['mocker']
+# Keep this authorship marker registry in sync across all pytest config roots.
+# Search for "agent_authored(model)" before editing.
+markers = [
+    "agent_authored(model): agent-authored test not yet materially human-reviewed",
+    "human_reviewed: agent-authored test materially reviewed or rewritten by a human",
+    "human_authored: test authored primarily by a human",
+]
 
 [tool.mypy]
 # Try to keep the mypy configuration similar between the subprojects
diff --git a/pytest.ini b/pytest.ini
index 9c11c2b5f5..76e5a2977d 100644
--- a/pytest.ini
+++ b/pytest.ini
@@ -20,6 +20,11 @@ markers =
     core: tests for cuda_core
     cython: cython tests
     smoke: meta-level smoke tests
+    # Keep this authorship marker registry in sync across all pytest config roots.
+    # Search for "agent_authored(model)" before editing.
+    agent_authored(model): agent-authored test not yet materially human-reviewed
+    human_reviewed: agent-authored test materially reviewed or rewritten by a human
+    human_authored: test authored primarily by a human
     flaky: mark test as flaky (provided by pytest-rerunfailures)
     # pytest-run-parallel related markers
     thread_unsafe: mark test as thread unsafe (provided by pytest-run-parallel)