Add meta-harness user-agent dimension for omnigent

Detect the OMNIGENT environment variable and report meta-harness/omnigent
as an independent user-agent dimension, parallel to agent detection. This
lets the platform see both the meta-harness (omnigent) and the underlying
coding agent (e.g. claude-code) in the same request.

Co-authored-by: Isaac
Signed-off-by: simon <simon.faltum@databricks.com>

Author: simonfaltum

NEXT_CHANGELOG.md

@@ -0,0 +1,19 @@
+# NEXT CHANGELOG
+
+## Release v0.118.0
+
+### New Features and Improvements
+
+* Added a `meta-harness` user-agent dimension that reports the omnigent meta-harness (detected via the `OMNIGENT` environment variable) independently of agent detection.
+
+### Security
+
+### Bug Fixes
+
+### Documentation
+
+### Breaking Changes
+
+### Internal Changes
+
+### API Changes

databricks/sdk/useragent.py

@@ -12,6 +12,7 @@
 RUNTIME_KEY = "runtime"
 CICD_KEY = "cicd"
 AUTH_KEY = "auth"
+META_HARNESS_KEY = "meta-harness"
 
 _product_name = "unknown"
 _product_version = "0.0.0"
@@ -175,6 +176,9 @@ def to_string(
     agent = agent_provider()
     if agent:
         base.append(("agent", agent))
+    meta_harness = meta_harness_provider()
+    if meta_harness:
+        base.append((META_HARNESS_KEY, meta_harness))
     return " ".join(f"{k}/{v}" for k, v in base)
 
 
@@ -329,3 +333,37 @@ def _agent_env_fallback() -> str:
     if not v:
         return ""
     return _sanitize_agent_value(v)[:_MAX_AGENT_FALLBACK_LEN]
+
+
+@dataclass(frozen=True)
+class _MetaHarnessRecord:
+    env_var: str
+    product: str
+
+
+# Known agent meta-harnesses, detected independently of agents (a meta-harness
+# is not an agent). Keep in sync with databricks-sdk-go and databricks-sdk-java.
+_KNOWN_META_HARNESSES: List[_MetaHarnessRecord] = [
+    _MetaHarnessRecord("OMNIGENT", "omnigent"),  # https://github.com/omnigent-ai/omnigent
+]
+
+# None = not computed, "" = computed but no meta-harness found.
+_meta_harness_provider = None
+
+
+def meta_harness_provider() -> str:
+    """Detect a known agent meta-harness by presence-only env var, else "".
+
+    Returns "multiple" if more than one matched. Cached after the first call.
+    """
+    global _meta_harness_provider
+    if _meta_harness_provider is not None:
+        return _meta_harness_provider
+    matches = [h.product for h in _KNOWN_META_HARNESSES if h.env_var in os.environ]
+    if len(matches) == 1:
+        _meta_harness_provider = matches[0]
+    elif len(matches) > 1:
+        _meta_harness_provider = "multiple"
+    else:
+        _meta_harness_provider = ""
+    return _meta_harness_provider

tests/test_config.py

@@ -82,6 +82,7 @@ def system(self):
     monkeypatch.setattr(useragent, "_extra", [])
     monkeypatch.setattr(useragent, "_cicd_provider", None)
     monkeypatch.setattr(useragent, "_agent_provider", None)
+    monkeypatch.setattr(useragent, "_meta_harness_provider", None)
 
     monkeypatch.setattr(platform, "python_version", lambda: "3.0.0")
     monkeypatch.setattr(platform, "uname", MockUname)

tests/test_core.py

@@ -275,6 +275,7 @@ def system(self):
     monkeypatch.setattr(useragent, "_extra", [])
     monkeypatch.setattr(useragent, "_cicd_provider", None)
     monkeypatch.setattr(useragent, "_agent_provider", None)
+    monkeypatch.setattr(useragent, "_meta_harness_provider", None)
 
     monkeypatch.setattr(platform, "python_version", lambda: "3.0.0")
     monkeypatch.setattr(platform, "uname", MockUname)

tests/test_user_agent.py

@@ -68,11 +68,12 @@ def clean_useragent_env():
     original_env = os.environ.copy()
     os.environ.clear()
 
-    # Clear cached CICD and agent providers.
+    # Clear cached CICD, agent, and meta-harness providers.
     from databricks.sdk import useragent
 
     useragent._cicd_provider = None
     useragent._agent_provider = None
+    useragent._meta_harness_provider = None
 
     yield
 
@@ -81,6 +82,7 @@ def clean_useragent_env():
     os.environ.update(original_env)
     useragent._cicd_provider = None
     useragent._agent_provider = None
+    useragent._meta_harness_provider = None
 
 
 def test_user_agent_cicd_no_provider(clean_useragent_env):
@@ -493,3 +495,64 @@ def test_agent_provider_cached(clean_useragent_env):
     os.environ["CLAUDECODE"] = "1"
 
     assert useragent.agent_provider() == "cursor"
+
+
+def test_meta_harness_provider_no_meta_harness(clean_useragent_env):
+    from databricks.sdk import useragent
+
+    assert useragent.meta_harness_provider() == ""
+
+
+def test_meta_harness_provider_omnigent(clean_useragent_env):
+    os.environ["OMNIGENT"] = "1"
+    from databricks.sdk import useragent
+
+    assert useragent.meta_harness_provider() == "omnigent"
+
+
+def test_meta_harness_provider_omnigent_empty_value_still_counts_as_set(clean_useragent_env):
+    # Presence-only matcher: an empty value still fires.
+    os.environ["OMNIGENT"] = ""
+    from databricks.sdk import useragent
+
+    assert useragent.meta_harness_provider() == "omnigent"
+
+
+def test_meta_harness_provider_cached(clean_useragent_env):
+    os.environ["OMNIGENT"] = "1"
+    from databricks.sdk import useragent
+
+    assert useragent.meta_harness_provider() == "omnigent"
+
+    # Change the environment: the cached result should persist.
+    del os.environ["OMNIGENT"]
+
+    assert useragent.meta_harness_provider() == "omnigent"
+
+
+def test_user_agent_string_includes_meta_harness(clean_useragent_env):
+    os.environ["OMNIGENT"] = "1"
+    from databricks.sdk import useragent
+
+    ua = useragent.to_string()
+    assert "meta-harness/omnigent" in ua
+
+
+def test_user_agent_string_no_meta_harness(clean_useragent_env):
+    from databricks.sdk import useragent
+
+    ua = useragent.to_string()
+    assert "meta-harness/" not in ua
+
+
+def test_meta_harness_independent_of_agent(clean_useragent_env):
+    # omnigent spawns the real agent CLI, so both env vars are set: the UA must
+    # carry both dimensions and omnigent must not trip the agent "multiple" signal.
+    os.environ["CLAUDECODE"] = "1"
+    os.environ["OMNIGENT"] = "1"
+    from databricks.sdk import useragent
+
+    ua = useragent.to_string()
+    assert "agent/claude-code" in ua
+    assert "meta-harness/omnigent" in ua
+    assert useragent.agent_provider() == "claude-code"