Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(proxy): reject Anthropic batch operations on Copilot (#3318)
Rod Boev871bbde37ae2
diff --git a/headroom/proxy/handlers/anthropic.py b/headroom/proxy/handlers/anthropic.pyindex c757ec9b7..7d67ce66f 100644--- a/headroom/proxy/handlers/anthropic.py+++ b/headroom/proxy/handlers/anthropic.py@@ -31,7 +31,11 @@ import httpxfrom headroom.agent_savings import proxy_pipeline_kwargsfrom headroom.ccr.context_tracker import looks_like_claude_code_compact_summaryfrom headroom.ccr.marker_resolution import resolve_markers_in_response-from headroom.copilot_auth import apply_copilot_api_auth, build_copilot_upstream_url+from headroom.copilot_auth import (+ apply_copilot_api_auth,+ build_copilot_upstream_url,+ is_copilot_upstream_url,+)from headroom.pipeline import PipelineStage, summarize_routing_markersfrom headroom.proxy.auth_mode import (classify_auth_mode,@@ -5062,6 +5066,26 @@ class AnthropicHandlerMixin:# permanently. The emit function is idempotent.await _finalize_pre_upstream()+ def _anthropic_batch_capability_error(self) -> Response | None:+ """Return the stable client error for a Copilot batch target."""+ if not is_copilot_upstream_url(self.ANTHROPIC_API_URL):+ return None++ from fastapi.responses import JSONResponse++ return JSONResponse(+ status_code=501,+ content={+ "type": "error",+ "error": {+ "type": "api_error",+ "message": (+ "Anthropic batch operations are not supported for Copilot targets."+ ),+ },+ },+ )+async def handle_anthropic_batch_create(self,request: Request,@@ -5087,6 +5111,9 @@ class AnthropicHandlerMixin:"""from fastapi.responses import JSONResponse, Response+ if (capability_error := self._anthropic_batch_capability_error()) is not None:+ return capability_error+from headroom.ccr import CCRToolInjectorfrom headroom.proxy.helpers import MAX_REQUEST_BODY_SIZE, _read_request_jsonfrom headroom.proxy.modes import is_cache_mode@@ -5426,6 +5453,9 @@ class AnthropicHandlerMixin:"""from fastapi.responses import Response+ if (capability_error := self._anthropic_batch_capability_error()) is not None:+ return capability_error+request_id = await self._next_request_id()start_time = time.time()path = request.url.path@@ -5566,6 +5596,9 @@ class AnthropicHandlerMixin:"""from fastapi.responses import Response+ if (capability_error := self._anthropic_batch_capability_error()) is not None:+ return capability_error+from headroom.ccr import BatchResultProcessor, get_batch_context_storerequest_id = await self._next_request_id()diff --git a/tests/test_copilot_auth.py b/tests/test_copilot_auth.pyindex 797b70568..70190b862 100644--- a/tests/test_copilot_auth.py+++ b/tests/test_copilot_auth.py@@ -901,6 +901,14 @@ def test_is_copilot_api_url_matches_ghe_copilot_hosts() -> None:assert not copilot_auth.is_copilot_api_url("https://not-copilot-api.acme.ghe.com/v1/responses")+def test_is_copilot_upstream_url_matches_public_and_enterprise_hosts() -> None:+ assert copilot_auth.is_copilot_upstream_url("https://api.githubcopilot.com/v1/messages/batches")+ assert copilot_auth.is_copilot_upstream_url(+ "https://copilot-api.acme.ghe.com/v1/messages/batches"+ )+ assert not copilot_auth.is_copilot_upstream_url("https://api.anthropic.com/v1/messages/batches")++def test_is_copilot_api_url_trusts_configured_enterprise_api_url(monkeypatch: pytest.MonkeyPatch,) -> None:diff --git a/tests/test_proxy_batch_integration.py b/tests/test_proxy_batch_integration.pyindex fa4bc33f0..36d9e7f92 100644--- a/tests/test_proxy_batch_integration.py+++ b/tests/test_proxy_batch_integration.py@@ -16,15 +16,16 @@ Run with:import jsonimport os+from unittest.mock import AsyncMockimport pytest+httpx = pytest.importorskip("httpx")pytest.importorskip("fastapi")-pytest.importorskip("httpx")-from fastapi.testclient import TestClient+from fastapi.testclient import TestClient # noqa: E402-from headroom.proxy.server import ProxyConfig, create_app+from headroom.proxy.server import ProxyConfig, create_app # noqa: E402# =============================================================================# Fixtures@@ -520,3 +521,134 @@ class TestBatchErrorHandling:content=b"not valid json",)assert response.status_code == 400+++def copilot_anthropic_batch_client():+ """Create a client whose resolved Anthropic target is public Copilot."""+ from headroom.proxy.server import HeadroomProxy++ original_anthropic_api_url = HeadroomProxy.ANTHROPIC_API_URL+ original_openai_api_url = HeadroomProxy.OPENAI_API_URL+ config = ProxyConfig(+ optimize=True,+ cache_enabled=False,+ rate_limit_enabled=False,+ cost_tracking_enabled=False,+ openai_api_url="https://api.githubcopilot.com",+ )+ app = create_app(config)+ with TestClient(app) as client:+ proxy = app.state.proxy+ proxy.http_client = AsyncMock()+ proxy.http_client.request.return_value = httpx.Response(+ 404,+ json={+ "type": "error",+ "error": {+ "type": "not_found_error",+ "message": "Batch endpoint not found",+ },+ },+ )+ try:+ yield client, proxy.http_client+ finally:+ HeadroomProxy.ANTHROPIC_API_URL = original_anthropic_api_url+ HeadroomProxy.OPENAI_API_URL = original_openai_api_url+++ ("method", "path", "kwargs"),+ [+ (+ "post",+ "/v1/messages/batches",+ {+ "json": {+ "requests": [+ {+ "custom_id": "req-1",+ "params": {+ "model": "claude-3-5-sonnet-20241022",+ "max_tokens": 1024,+ "messages": [{"role": "user", "content": "Hello"}],+ },+ }+ ]+ }+ },+ ),+ ("get", "/v1/messages/batches", {}),+ ("get", "/v1/messages/batches/batch_123", {}),+ ("post", "/v1/messages/batches/batch_123/cancel", {}),+ ("get", "/v1/messages/batches/batch_123/results", {}),+ ],+)+def test_copilot_anthropic_batch_routes_are_rejected_before_upstream(+ copilot_anthropic_batch_client, method, path, kwargs+):+ """Every registered batch route returns one stable capability response."""+ client, http_client = copilot_anthropic_batch_client++ response = getattr(client, method)(path, **kwargs)++ assert response.status_code == 501+ assert response.json() == {+ "type": "error",+ "error": {+ "type": "api_error",+ "message": "Anthropic batch operations are not supported for Copilot targets.",+ },+ }+ http_client.request.assert_not_called()+ http_client.get.assert_not_called()+ if method == "post":+ http_client.post.assert_not_called()+++def test_copilot_anthropic_batch_rejection_precedes_body_parsing(copilot_anthropic_batch_client):+ """Capability rejection wins even when the batch body is invalid."""+ client, http_client = copilot_anthropic_batch_client++ response = client.post("/v1/messages/batches", content=b"not json")++ assert response.status_code == 501+ assert response.json()["error"]["type"] == "api_error"+ http_client.request.assert_not_called()+ http_client.post.assert_not_called()+++def test_explicit_non_copilot_anthropic_target_forwards():+ """An explicit Anthropic target remains outside the Copilot guard."""+ from headroom.proxy.server import HeadroomProxy++ original_anthropic_api_url = HeadroomProxy.ANTHROPIC_API_URL+ original_openai_api_url = HeadroomProxy.OPENAI_API_URL+ config = ProxyConfig(+ optimize=False,+ cache_enabled=False,+ rate_limit_enabled=False,+ cost_tracking_enabled=False,+ openai_api_url="https://api.githubcopilot.com",+ anthropic_api_url="https://api.anthropic.com",+ )+ app = create_app(config)+ try:+ with TestClient(app) as client:+ proxy = app.state.proxy+ proxy.http_client = AsyncMock()+ proxy.http_client.request.return_value = httpx.Response(200, json={"data": []})+ http_client = proxy.http_client++ response = client.get("/v1/messages/batches")+ finally:+ HeadroomProxy.ANTHROPIC_API_URL = original_anthropic_api_url+ HeadroomProxy.OPENAI_API_URL = original_openai_api_url++ assert response.status_code == 200+ http_client.request.assert_called_once()+ assert http_client.request.call_args.kwargs["url"] == (+ "https://api.anthropic.com/v1/messages/batches"+ )+ http_client.post.assert_not_called()