feat(tasks): Accept context window and fast mode on runs

frontend/snapshots.yml

@@ -5395,13 +5395,13 @@ snapshots:
     scenes-app-customer-analytics-dashboard--b-2-c-mode--light:
         hash: v1.k794b7964.ff25e11868464e00df352180254dbe0a1017cd1d8c7c4f4c7d6995068e80e2dc.m0XFOZVPCdSaCLaPQ6XTePvkZJ1I11QI2KARZi9QpWs
     scenes-app-customer-analytics-dashboard--gated-with-feature-toggle--dark:
-        hash: v1.k794b7964.f07be6c5a9a602a1b2de2e07ff8329cf79022a86359c8592ec0ffa997e06815e.4p0ObyV8dflPFp4eV8-iGtIyPw-H4rNRImtZ98MUaq0
+        hash: v1.k794b7964.731394bccab76141d6dc494f99dd56cc85e18b3f787977f8edd5c1a50c5564c8.9z2bNOo-dAM0tEyxcqStccQSV7X6iHbKnJdqAgvNapU
     scenes-app-customer-analytics-dashboard--gated-with-feature-toggle--light:
-        hash: v1.k794b7964.c08ac439ce99c1885784bb01585643983c42e334ee9b66dc4fdca3a200f49aa8.1nfQJfx8u9E46SodXwq0FKkEXnMgRERGpZ3LrS7lsgk
+        hash: v1.k794b7964.d4a3f7e666d9788b86e4e35e6e976dc048f50189129b316352f60e389916e740.XNraHFrJz2oo-ioKjtSG3b53aSd7SUozc4GkGnRUVPw
     scenes-app-customer-analytics-dashboard--gated-without-matching-early-access-feature--dark:
-        hash: v1.k794b7964.df78ad055d2df40bd8f7291c6be67f5186e055f66f8d642e54f4493bfea8dee4.ZN0UYUnaFqZlJQb_8eCa8a74QdPPt4v9VBEQNfMT5Dw
+        hash: v1.k794b7964.212105add8f003ebda9493c4f1324aa86a9ea9d4f078a9202f746125bc188ee0.mcCJ4cjTJAjHr9-eqJV3Vki9l2-kPcGcbCkwbnXGGNI
     scenes-app-customer-analytics-dashboard--gated-without-matching-early-access-feature--light:
-        hash: v1.k794b7964.261496c85d5d04af3eb5e766063ec3e1783ea9ba2d54f82bb9c7d4b0f1cc28a0.5Gacg7ZDoAp-d35eiCG9-Jp2vYDw-4uvGAJsL5U4uiI
+        hash: v1.k794b7964.c17e8855ba266d7d7732b79d59cbe6996e34469b95aaf820146a2ec098d4c1cd.xHHIXeNzT7uaxywPDDglLPD96SXg9n66LgqmXi677J0
     scenes-app-customer-analytics-journey-builder--new-journey--dark:
         hash: v1.k794b7964.5e2f6c51e7e4db5d5d8bed07edeb90f8e71b6dbe9e4aa7042ce9337e407ff3bb.qbiM7NeFkMfNbY374tsbc_nZloshk9lmHLFgIPx5-ow
     scenes-app-customer-analytics-journey-builder--new-journey--light:

products/posthog_ai/frontend/utils/composerModels.ts

@@ -27,6 +27,7 @@ const EFFORT_LABELS: Record<ReasoningEffortEnumApi, string> = {
     [ReasoningEffortEnumApi.High]: 'High',
     [ReasoningEffortEnumApi.Xhigh]: 'Extra high',
     [ReasoningEffortEnumApi.Max]: 'Max',
+    [ReasoningEffortEnumApi.Ultracode]: 'Ultracode',
 }
 
 // Mirrors backend CLAUDE_REASONING_EFFORTS_BY_MODEL (products/tasks/backend/temporal/process_task/utils.py):
@@ -38,15 +39,24 @@ const EFFORTS_BY_MODEL: Record<string, ReasoningEffortEnumApi[]> = {
         ReasoningEffortEnumApi.High,
         ReasoningEffortEnumApi.Xhigh,
         ReasoningEffortEnumApi.Max,
+        ReasoningEffortEnumApi.Ultracode,
     ],
     'claude-opus-5': [
         ReasoningEffortEnumApi.Low,
         ReasoningEffortEnumApi.Medium,
         ReasoningEffortEnumApi.High,
         ReasoningEffortEnumApi.Xhigh,
         ReasoningEffortEnumApi.Max,
+        ReasoningEffortEnumApi.Ultracode,
+    ],
+    'claude-sonnet-5': [
+        ReasoningEffortEnumApi.Low,
+        ReasoningEffortEnumApi.Medium,
+        ReasoningEffortEnumApi.High,
+        ReasoningEffortEnumApi.Xhigh,
+        ReasoningEffortEnumApi.Max,
+        ReasoningEffortEnumApi.Ultracode,
     ],
-    'claude-sonnet-5': [ReasoningEffortEnumApi.Low, ReasoningEffortEnumApi.Medium, ReasoningEffortEnumApi.High],
 }
 
 const FALLBACK_EFFORTS: ReasoningEffortEnumApi[] = [

products/tasks/backend/facade/api.py

@@ -3086,6 +3086,8 @@ def bootstrap_task_run(
     runtime_adapter = validated_data.get("runtime_adapter")
     model = validated_data.get("model")
     reasoning_effort = validated_data.get("reasoning_effort")
+    context_window = validated_data.get("context_window")
+    fast_mode = validated_data.get("fast_mode")
     github_user_token = validated_data.get("github_user_token")
     initial_permission_mode = validated_data.get("initial_permission_mode")
     imported_mcp_servers = validated_data.get("imported_mcp_servers")
@@ -3108,6 +3110,8 @@ def bootstrap_task_run(
         "provider": provider,
         "model": model,
         "reasoning_effort": reasoning_effort,
+        "context_window": context_window,
+        "fast_mode": fast_mode,
         "rtk_enabled": validated_data.get("rtk_enabled"),
     }.items():
         if value is not None:
@@ -4399,12 +4403,16 @@ def run_task(
                 warm_state.get("runtime_adapter") or None,
                 warm_state.get("model") or None,
                 warm_state.get("reasoning_effort") or None,
+                warm_state.get("context_window") or None,
+                warm_state.get("fast_mode") or None,
                 warm_state.get("sandbox_environment_id") or None,
                 warm_state.get("custom_image_id") or None,
             ) == (
                 validated_data.get("runtime_adapter") or None,
                 validated_data.get("model") or None,
                 validated_data.get("reasoning_effort") or None,
+                validated_data.get("context_window") or None,
+                validated_data.get("fast_mode") or None,
                 str(validated_data["sandbox_environment_id"]) if validated_data.get("sandbox_environment_id") else None,
                 str(validated_data["custom_image_id"]) if validated_data.get("custom_image_id") else None,
             )
@@ -4448,6 +4456,8 @@ def run_task(
     runtime_adapter = validated_data.get("runtime_adapter")
     model = validated_data.get("model")
     reasoning_effort = validated_data.get("reasoning_effort")
+    context_window = validated_data.get("context_window")
+    fast_mode = validated_data.get("fast_mode")
     github_user_token = validated_data.get("github_user_token")
     initial_permission_mode = validated_data.get("initial_permission_mode")
     imported_mcp_servers = validated_data.get("imported_mcp_servers")
@@ -4463,6 +4473,8 @@ def run_task(
         "runtime_adapter": runtime_adapter,
         "model": model,
         "reasoning_effort": reasoning_effort,
+        "context_window": context_window,
+        "fast_mode": fast_mode,
     }
 
     extra_state: dict | None = None
@@ -4521,6 +4533,8 @@ def run_task(
         runtime_adapter = runtime_state_fields["runtime_adapter"]
         model = runtime_state_fields["model"]
         reasoning_effort = runtime_state_fields["reasoning_effort"]
+        context_window = runtime_state_fields["context_window"]
+        fast_mode = runtime_state_fields["fast_mode"]
         if branch is None and prev_state.pr_base_branch is not None:
             branch = prev_state.pr_base_branch
 
@@ -4536,6 +4550,8 @@ def run_task(
         "provider": provider,
         "model": model,
         "reasoning_effort": reasoning_effort,
+        "context_window": context_window,
+        "fast_mode": fast_mode,
     }.items():
         if value is not None:
             extra_state = extra_state or {}

products/tasks/backend/facade/run_config.py

@@ -21,6 +21,7 @@
 # so presentation builds serializer choices without importing the ORM model directly.
 from products.tasks.backend.models import TaskArtifact as _TaskArtifact
 from products.tasks.backend.temporal.process_task.utils import (
+    CONTEXT_WINDOW_CHOICES,
     PUBLIC_REASONING_EFFORTS,
     GitHubCredentialSource,
     LLMProvider,
@@ -44,6 +45,7 @@
 __all__ = [
     "ALL_INITIAL_PERMISSION_MODE_CHOICES",
     "CODEX_INITIAL_PERMISSION_MODE_CHOICES",
+    "CONTEXT_WINDOW_CHOICES",
     "INITIAL_PERMISSION_MODE_CHOICES",
     "InitialPermissionMode",
     "PUBLIC_REASONING_EFFORTS",

products/tasks/backend/logic/services/docker_sandbox.py

@@ -794,6 +794,8 @@ def _build_agent_server_command(
         provider: str | None = None,
         model: str | None = None,
         reasoning_effort: str | None = None,
+        context_window: str | None = None,
+        fast_mode: bool | None = None,
         initial_permission_mode: str | None = None,
         mcp_servers_arg: str = "",
         relay_mcp_servers_arg: str = "",
@@ -815,6 +817,8 @@ def _build_agent_server_command(
             provider=provider,
             model=model,
             reasoning_effort=reasoning_effort,
+            context_window=context_window,
+            fast_mode=fast_mode,
             initial_permission_mode=initial_permission_mode,
             event_ingest_token=event_ingest_token,
             event_ingest_url=event_ingest_url,
@@ -899,6 +903,8 @@ def start_agent_server(
         provider: str | None = None,
         model: str | None = None,
         reasoning_effort: str | None = None,
+        context_window: str | None = None,
+        fast_mode: bool | None = None,
         initial_permission_mode: str | None = None,
         mcp_configs: list[McpServerConfig] | None = None,
         relayed_mcp_servers: list[str] | None = None,
@@ -970,9 +976,11 @@ def start_agent_server(
             provider,
             model,
             reasoning_effort,
-            initial_permission_mode,
-            mcp_servers_arg,
-            relay_mcp_servers_arg,
+            context_window=context_window,
+            fast_mode=fast_mode,
+            initial_permission_mode=initial_permission_mode,
+            mcp_servers_arg=mcp_servers_arg,
+            relay_mcp_servers_arg=relay_mcp_servers_arg,
             allowed_domains=allowed_domains,
             event_ingest_token=event_ingest_token,
             event_ingest_url=event_ingest_url,
@@ -1021,6 +1029,8 @@ def start_agent_server(
                 provider=provider,
                 model=model,
                 reasoning_effort=reasoning_effort,
+                context_window=context_window,
+                fast_mode=fast_mode,
                 initial_permission_mode=initial_permission_mode,
                 mcp_servers_arg=mcp_servers_arg,
                 relay_mcp_servers_arg=relay_mcp_servers_arg,

products/tasks/backend/logic/services/modal_sandbox.py

@@ -949,6 +949,8 @@ def _build_agent_server_command(
         provider: str | None = None,
         model: str | None = None,
         reasoning_effort: str | None = None,
+        context_window: str | None = None,
+        fast_mode: bool | None = None,
         initial_permission_mode: str | None = None,
         mcp_servers_arg: str = "",
         relay_mcp_servers_arg: str = "",
@@ -966,6 +968,8 @@ def _build_agent_server_command(
             provider=provider,
             model=model,
             reasoning_effort=reasoning_effort,
+            context_window=context_window,
+            fast_mode=fast_mode,
             initial_permission_mode=initial_permission_mode,
             event_ingest_token=event_ingest_token,
             event_ingest_url=event_ingest_url,
@@ -1076,6 +1080,8 @@ def start_agent_server(
         provider: str | None = None,
         model: str | None = None,
         reasoning_effort: str | None = None,
+        context_window: str | None = None,
+        fast_mode: bool | None = None,
         initial_permission_mode: str | None = None,
         mcp_configs: list[McpServerConfig] | None = None,
         relayed_mcp_servers: list[str] | None = None,
@@ -1150,9 +1156,11 @@ def start_agent_server(
             provider,
             model,
             reasoning_effort,
-            initial_permission_mode,
-            mcp_servers_arg,
-            relay_mcp_servers_arg,
+            context_window=context_window,
+            fast_mode=fast_mode,
+            initial_permission_mode=initial_permission_mode,
+            mcp_servers_arg=mcp_servers_arg,
+            relay_mcp_servers_arg=relay_mcp_servers_arg,
             allowed_domains=allowed_domains,
             event_ingest_token=event_ingest_token,
             event_ingest_url=event_ingest_url,

products/tasks/backend/logic/services/sandbox.py

@@ -171,6 +171,8 @@ def build_agent_runtime_env_prefix(
     provider: str | None = None,
     model: str | None = None,
     reasoning_effort: str | None = None,
+    context_window: str | None = None,
+    fast_mode: bool | None = None,
     initial_permission_mode: str | None = None,
     event_ingest_token: str | None = None,
     event_ingest_url: str | None = None,
@@ -183,6 +185,9 @@ def build_agent_runtime_env_prefix(
         "POSTHOG_CODE_PROVIDER": provider,
         "POSTHOG_CODE_MODEL": model,
         "POSTHOG_CODE_REASONING_EFFORT": reasoning_effort,
+        "POSTHOG_CODE_CONTEXT_WINDOW": context_window,
+        # Explicit false pins fast mode off even if a stale env value survives in a resumed sandbox.
+        "POSTHOG_CODE_FAST_MODE": None if fast_mode is None else ("true" if fast_mode else "false"),
         "POSTHOG_CODE_INITIAL_PERMISSION_MODE": initial_permission_mode,
         "POSTHOG_TASK_RUN_EVENT_INGEST_TOKEN": event_ingest_token,
         "POSTHOG_TASK_RUN_EVENT_INGEST_URL": event_ingest_url,
@@ -333,6 +338,8 @@ def start_agent_server(
         provider: str | None = None,
         model: str | None = None,
         reasoning_effort: str | None = None,
+        context_window: str | None = None,
+        fast_mode: bool | None = None,
         initial_permission_mode: str | None = None,
         mcp_configs: list[McpServerConfig] | None = None,
         relayed_mcp_servers: list[str] | None = None,

products/tasks/backend/logic/services/test_docker_sandbox.py

@@ -590,6 +590,8 @@ def test_start_agent_server_includes_runtime_environment_variables(self):
                     provider="openai",
                     model="gpt-5.3-codex",
                     reasoning_effort="medium",
+                    context_window="1m",
+                    fast_mode=True,
                     initial_permission_mode="plan",
                     event_ingest_token="ingest-token",
                     event_ingest_url="http://localhost:8003",
@@ -600,11 +602,44 @@ def test_start_agent_server_includes_runtime_environment_variables(self):
         assert "POSTHOG_CODE_PROVIDER=openai" in command
         assert "POSTHOG_CODE_MODEL=gpt-5.3-codex" in command
         assert "POSTHOG_CODE_REASONING_EFFORT=medium" in command
+        assert "POSTHOG_CODE_CONTEXT_WINDOW=1m" in command
+        assert "POSTHOG_CODE_FAST_MODE=true" in command
         assert "POSTHOG_CODE_INITIAL_PERMISSION_MODE=plan" in command
         assert "POSTHOG_TASK_RUN_EVENT_INGEST_TOKEN=ingest-token" in command
         # The host proxy URL is rewritten so it resolves from inside the container.
         assert "POSTHOG_TASK_RUN_EVENT_INGEST_URL=http://host.docker.internal:8003" in command
 
+    @pytest.mark.parametrize(
+        "fast_mode, expected_env",
+        [
+            (False, "POSTHOG_CODE_FAST_MODE=false"),
+            (None, None),
+        ],
+    )
+    def test_start_agent_server_fast_mode_env(self, fast_mode, expected_env):
+        sandbox = DockerSandbox.__new__(DockerSandbox)
+        sandbox._container_id = "abc123"
+        sandbox.id = "abc123"
+        sandbox.config = SandboxConfig(name="test")
+        sandbox._host_port = 12345
+
+        with patch.object(sandbox, "is_running", return_value=True):
+            with patch.object(sandbox, "execute") as mock_execute:
+                mock_execute.return_value = ExecutionResult(stdout="ok:1", stderr="", exit_code=0, error=None)
+                sandbox.start_agent_server(
+                    "posthog/posthog",
+                    "task-123",
+                    "run-456",
+                    "background",
+                    fast_mode=fast_mode,
+                )
+
+        command = _agent_server_launch_command(mock_execute)
+        if expected_env is not None:
+            assert expected_env in command
+        else:
+            assert "POSTHOG_CODE_FAST_MODE" not in command
+
     @pytest.mark.parametrize(
         "keep_stream_open, expected_env_present",
         [

products/tasks/backend/logic/services/tests/test_modal_sandbox.py

@@ -549,6 +549,8 @@ def test_start_agent_server_includes_runtime_environment_variables(self, mock_sa
             provider="openai",
             model="gpt-5.3-codex",
             reasoning_effort="high",
+            context_window="1m",
+            fast_mode=True,
             initial_permission_mode="plan",
             event_ingest_token="ingest-token",
             event_ingest_url="https://agent-proxy.example.com",
@@ -559,12 +561,40 @@ def test_start_agent_server_includes_runtime_environment_variables(self, mock_sa
         assert "POSTHOG_CODE_PROVIDER=openai" in command
         assert "POSTHOG_CODE_MODEL=gpt-5.3-codex" in command
         assert "POSTHOG_CODE_REASONING_EFFORT=high" in command
+        assert "POSTHOG_CODE_CONTEXT_WINDOW=1m" in command
+        assert "POSTHOG_CODE_FAST_MODE=true" in command
         assert "POSTHOG_CODE_INITIAL_PERMISSION_MODE=plan" in command
         assert "POSTHOG_TASK_RUN_EVENT_INGEST_TOKEN=ingest-token" in command
         # Modal sandboxes reach the proxy by its real URL, no Docker-host rewrite.
         assert "POSTHOG_TASK_RUN_EVENT_INGEST_URL=https://agent-proxy.example.com" in command
         assert "POSTHOG_RTK=1" in command
 
+    @pytest.mark.parametrize(
+        "fast_mode, expected_env",
+        [
+            (False, "POSTHOG_CODE_FAST_MODE=false"),
+            (None, None),
+        ],
+    )
+    def test_start_agent_server_fast_mode_env(self, mock_sandbox: Any, fast_mode, expected_env):
+        mock_sandbox.execute = MagicMock(
+            return_value=ExecutionResult(stdout="ok:1", stderr="", exit_code=0, error=None),
+        )
+
+        mock_sandbox.start_agent_server(
+            repository="posthog/posthog",
+            task_id="task-123",
+            run_id="run-456",
+            mode="background",
+            fast_mode=fast_mode,
+        )
+
+        command = _agent_server_launch_command(mock_sandbox.execute)
+        if expected_env is not None:
+            assert expected_env in command
+        else:
+            assert "POSTHOG_CODE_FAST_MODE" not in command
+
     @pytest.mark.parametrize(
         "rtk_enabled, expected_env",
         [