plm-engine-core 1.3.0__tar.gz → 1.4.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/PKG-INFO +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/__init__.py +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/dto.py +76 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/executor.py +44 -5
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/registry.py +16 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/main.py +47 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/routers/agent_runs.py +10 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/settings.py +10 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/agent_telemetry_service.py +18 -2
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/invocation_telemetry.py +9 -2
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/llm_service.py +9 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/PKG-INFO +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/SOURCES.txt +2 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/pyproject.toml +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_turn_execution.py +2 -2
- plm_engine_core-1.4.0/tests/test_dev_tenant_middleware.py +110 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_execution_error_normalization.py +3 -3
- plm_engine_core-1.4.0/tests/test_governed_turn_contract_extension.py +286 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_mcrc_trace.py +2 -2
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_proof_driver.py +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_skill_dispatch_turn.py +1 -1
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/README.md +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/agent_runtime/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/agent_runtime/dispatcher.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/agent_runtime/selector.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/cli/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/cli/__main__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/cli/auth.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/_lint_fixtures/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/_lint_fixtures/violation_demo.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/capability_registry/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/capability_registry/errors.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/capability_registry/registry.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/connectors/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/connectors/dispatcher.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/connectors/runtime.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/dispatch.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/hitl/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/hitl/errors.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/hitl/gate.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/hitl/sql_store.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/identity/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/identity/middleware.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/identity/provider.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/identity/system_identity.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/identity/telemetry.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/knowledge/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/knowledge/curation_index.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/knowledge/envelope.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/knowledge/runtime.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/authorization.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/autonomy_gating.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/base.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/cost_budget.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/policies/rate_limit.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/retry_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/control_plane/run_task_tracker.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/__main__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/bundle.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/event_bus.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/governance.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/hitl.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/lifecycle.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/mcrc.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/run_projection.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/snapshot.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/taxonomy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/readiness.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/routers/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/routers/config.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/routers/health.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/llm_catalog/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/llm_catalog/loader.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/llm_catalog/model_list.yaml +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/asgi.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/errors.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/manifest.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/openapi.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/server.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/mcp/tools.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/database.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/database_service.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/default_bindings.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/invocation_context.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/invocation_lifecycle.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/llm_env.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/plan_analysis_service.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/pricing_service.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/retry_orchestrator.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/retry_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/run_task_tracker.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/scoped_state.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/token_estimator.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/workspace_port.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/studio/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/studio/readers/__init__.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/studio/readers/mcrc_v1.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/dependency_links.txt +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/entry_points.txt +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/requires.txt +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core.egg-info/top_level.txt +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/setup.cfg +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_run_contract.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_run_feature_flag.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_run_hitl_contract.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_run_mcrc_projection.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_agent_snapshot.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_authorization_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_autonomy_gating_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_capability_registry.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_cli_auth.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_connector_dispatcher.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_contract_bundle.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_core_host_build_app.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_core_host_wave9_parity_smoke.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_cost_budget_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_error_code_mirror.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_error_taxonomy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_evidence_pack.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_evidence_pack_cli.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_g8_conformance.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_gate_verdict.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_governance_chain.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_health_readiness.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_hitl_gate.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_hitl_resume.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_identity_middleware.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_identity_provider.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_import_linter.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_imports.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_knowledge_curation_index.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_knowledge_envelope.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_knowledge_runtime_adapter.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_mcp_asgi.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_mcp_server.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_mcrc_emitter_wiring.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_mcrc_v1_reader.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_model_dispatch.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_openapi.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_pf012_measure.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_plm_skill_kernel_optional.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_policy_pipeline_order.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_promotion_gate.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_rate_limit_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_retry_policy.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_run_lifecycle.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_run_task_tracker.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_skill_selector.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_sse_backpressure.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_sse_event_stream.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_sse_reconnect.py +0 -0
- {plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/tests/test_system_identity.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: plm-engine-core
|
|
3
|
-
Version: 1.
|
|
3
|
+
Version: 1.4.0
|
|
4
4
|
Summary: TracePulse PLM Engine Core — control_plane (governance, autonomy, HITL, run lifecycle, trace propagation) + agent_runtime (selector, dispatcher, retry, escalation). US-CR.1 scaffold; behaviour lands in US-CR.0+. Internal one-way import contract: control_plane MUST NOT import from agent_runtime (mechanical enforcement in US-CR.2).
|
|
5
5
|
Author: TracePulse
|
|
6
6
|
License: Proprietary
|
|
@@ -97,6 +97,52 @@ class AgentSnapshot(BaseModel):
|
|
|
97
97
|
description="Skill references by id (an agent is not a skill — INV-003).",
|
|
98
98
|
max_length=64,
|
|
99
99
|
)
|
|
100
|
+
# ADR-0005 D1 — the agent's persona/expertise instruction. Placed on the
|
|
101
|
+
# snapshot (not the request) because it *is* agent configuration: it is
|
|
102
|
+
# therefore digest-covered and secret-scanned by E4 automatically
|
|
103
|
+
# (``_digest_content`` dumps the whole model minus ``digest``). Optional, so
|
|
104
|
+
# a consumer that omits it keeps the previous governed-runtime default.
|
|
105
|
+
system_prompt: Optional[str] = Field(
|
|
106
|
+
default=None,
|
|
107
|
+
max_length=65536,
|
|
108
|
+
description=(
|
|
109
|
+
"Agent persona/expertise system instruction (secret-free, "
|
|
110
|
+
"digest-covered). Omit to use the governed runtime default."
|
|
111
|
+
),
|
|
112
|
+
)
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
class TurnRole(str, Enum):
|
|
116
|
+
"""Role of a replayed conversational message (ADR-0005 D1)."""
|
|
117
|
+
|
|
118
|
+
USER = "user"
|
|
119
|
+
ASSISTANT = "assistant"
|
|
120
|
+
SYSTEM = "system"
|
|
121
|
+
|
|
122
|
+
|
|
123
|
+
class TurnMessage(BaseModel):
|
|
124
|
+
"""One prior conversational turn replayed into a governed run (ADR-0005 D1).
|
|
125
|
+
|
|
126
|
+
Turn-scoped, not agent configuration — so this travels on the *request*,
|
|
127
|
+
never on the snapshot, and never enters the snapshot digest.
|
|
128
|
+
"""
|
|
129
|
+
|
|
130
|
+
model_config = ConfigDict(extra="forbid")
|
|
131
|
+
|
|
132
|
+
role: TurnRole
|
|
133
|
+
content: str = Field(min_length=1, max_length=65536)
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
class TurnMode(str, Enum):
|
|
137
|
+
"""Conversational mode of a governed turn (ADR-0005 D1).
|
|
138
|
+
|
|
139
|
+
Mirrors the consumer-side agent/plan/ask modes. Optional on the request; a
|
|
140
|
+
run without an explicit mode is executed as ``agent``.
|
|
141
|
+
"""
|
|
142
|
+
|
|
143
|
+
AGENT = "agent"
|
|
144
|
+
PLAN = "plan"
|
|
145
|
+
ASK = "ask"
|
|
100
146
|
|
|
101
147
|
|
|
102
148
|
class RunContext(BaseModel):
|
|
@@ -129,6 +175,21 @@ class AgentRunCreateRequest(BaseModel):
|
|
|
129
175
|
agent: AgentSnapshot
|
|
130
176
|
message: str = Field(min_length=1)
|
|
131
177
|
context: Optional[RunContext] = None
|
|
178
|
+
# ADR-0005 D1 — turn-scoped conversational inputs. Both optional, so a
|
|
179
|
+
# single-shot consumer is unaffected. ``history`` is a bounded replay window
|
|
180
|
+
# the consumer chooses; Core does not read a conversation store.
|
|
181
|
+
history: Optional[List[TurnMessage]] = Field(
|
|
182
|
+
default=None,
|
|
183
|
+
max_length=200,
|
|
184
|
+
description=(
|
|
185
|
+
"Bounded prior-turn replay window, oldest first. Core is stateless "
|
|
186
|
+
"w.r.t. the consumer's conversation store (parity row P2)."
|
|
187
|
+
),
|
|
188
|
+
)
|
|
189
|
+
mode: Optional[TurnMode] = Field(
|
|
190
|
+
default=None,
|
|
191
|
+
description="Conversational mode; absent is executed as 'agent'.",
|
|
192
|
+
)
|
|
132
193
|
|
|
133
194
|
|
|
134
195
|
# ── Responses ─────────────────────────────────────────────────────────────
|
|
@@ -160,6 +221,18 @@ class AgentRunResource(BaseModel):
|
|
|
160
221
|
state: RunState
|
|
161
222
|
links: RunLinks
|
|
162
223
|
created_at: str = Field(description="ISO-8601 UTC creation timestamp.")
|
|
224
|
+
# ADR-0005 D1 — the turn's reply text. Before this field the assistant's
|
|
225
|
+
# output had no channel out of Core at all: it lived only on the internal
|
|
226
|
+
# ``ExecutionResult`` for the E6 trace handoff. Optional and non-required,
|
|
227
|
+
# so a consumer pinned to the pre-extension shape stays wire-compatible.
|
|
228
|
+
# ``None`` until a terminal state carries a provider reply.
|
|
229
|
+
output_text: Optional[str] = Field(
|
|
230
|
+
default=None,
|
|
231
|
+
description=(
|
|
232
|
+
"Assistant reply text for a terminal run; None while non-terminal, "
|
|
233
|
+
"on failure, or when the turn required no provider call."
|
|
234
|
+
),
|
|
235
|
+
)
|
|
163
236
|
|
|
164
237
|
|
|
165
238
|
__all__ = [
|
|
@@ -168,6 +241,9 @@ __all__ = [
|
|
|
168
241
|
"GenerationParameters",
|
|
169
242
|
"ModelConfig",
|
|
170
243
|
"AgentSnapshot",
|
|
244
|
+
"TurnRole",
|
|
245
|
+
"TurnMessage",
|
|
246
|
+
"TurnMode",
|
|
171
247
|
"RunContext",
|
|
172
248
|
"AgentRunCreateRequest",
|
|
173
249
|
"RunLinks",
|
{plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/executor.py
RENAMED
|
@@ -61,31 +61,60 @@ class LlmTurnResult:
|
|
|
61
61
|
total_tokens: Optional[int] = None
|
|
62
62
|
|
|
63
63
|
|
|
64
|
+
#: Governed runtime default used when a snapshot announces no ``system_prompt``.
|
|
65
|
+
DEFAULT_SYSTEM_PROMPT = "You are a governed PLM agent runtime."
|
|
66
|
+
|
|
67
|
+
|
|
64
68
|
class LlmTurnCallable(Protocol):
|
|
65
69
|
"""A single governed LLM turn. The concrete model + parameters are the
|
|
66
70
|
authorized/resolved values; credentials are resolved runtime-side only and
|
|
67
|
-
never passed through this seam.
|
|
71
|
+
never passed through this seam.
|
|
72
|
+
|
|
73
|
+
``system_prompt``/``history``/``mode`` arrive from the ADR-0005 D1 contract
|
|
74
|
+
fields. All three are optional: a single-shot consumer that omits them gets
|
|
75
|
+
exactly the pre-extension behaviour.
|
|
76
|
+
"""
|
|
68
77
|
|
|
69
78
|
def __call__(
|
|
70
|
-
self,
|
|
79
|
+
self,
|
|
80
|
+
*,
|
|
81
|
+
model: str,
|
|
82
|
+
message: str,
|
|
83
|
+
parameters: Mapping[str, Any],
|
|
84
|
+
system_prompt: Optional[str] = None,
|
|
85
|
+
history: Optional["list[Mapping[str, str]]"] = None,
|
|
86
|
+
mode: Optional[str] = None,
|
|
71
87
|
) -> Awaitable[LlmTurnResult]: ...
|
|
72
88
|
|
|
73
89
|
|
|
74
90
|
async def _production_llm_call(
|
|
75
|
-
*,
|
|
91
|
+
*,
|
|
92
|
+
model: str,
|
|
93
|
+
message: str,
|
|
94
|
+
parameters: Mapping[str, Any],
|
|
95
|
+
system_prompt: Optional[str] = None,
|
|
96
|
+
history: Optional["list[Mapping[str, str]]"] = None,
|
|
97
|
+
mode: Optional[str] = None,
|
|
76
98
|
) -> LlmTurnResult:
|
|
77
99
|
"""Default provider adapter over :class:`LlmService` (lazy import so local
|
|
78
|
-
tests never require provider credentials or the LiteLLM import chain).
|
|
100
|
+
tests never require provider credentials or the LiteLLM import chain).
|
|
101
|
+
|
|
102
|
+
``mode`` is **not** semantically reinterpreted here: the consumer owns
|
|
103
|
+
mode-specific prompting and encodes it in ``system_prompt``. It travels this
|
|
104
|
+
seam so it is governed and traced, and so an alternative adapter can honour
|
|
105
|
+
it (ADR-0005 D1).
|
|
106
|
+
"""
|
|
79
107
|
from ...runtime.llm_service import llm_service # lazy singleton
|
|
80
108
|
|
|
81
109
|
temperature = parameters.get("temperature")
|
|
82
110
|
max_tokens = parameters.get("max_output_tokens")
|
|
83
111
|
text, usage = await llm_service.generate_reply_with_usage(
|
|
84
|
-
system_prompt=
|
|
112
|
+
system_prompt=system_prompt or DEFAULT_SYSTEM_PROMPT,
|
|
85
113
|
user_prompt=message,
|
|
86
114
|
temperature=temperature if temperature is not None else 0.7,
|
|
87
115
|
model=model,
|
|
88
116
|
max_tokens=max_tokens,
|
|
117
|
+
history=list(history) if history else None,
|
|
89
118
|
)
|
|
90
119
|
return LlmTurnResult(
|
|
91
120
|
text=text,
|
|
@@ -487,6 +516,9 @@ async def execute_turn(
|
|
|
487
516
|
autonomy_level: str = DEFAULT_AUTONOMY_LEVEL,
|
|
488
517
|
parameters: Optional[Mapping[str, Any]] = None,
|
|
489
518
|
skills: Optional["list[str]"] = None,
|
|
519
|
+
system_prompt: Optional[str] = None,
|
|
520
|
+
history: Optional["list[Mapping[str, str]]"] = None,
|
|
521
|
+
mode: Optional[str] = None,
|
|
490
522
|
tenant_id: Optional[str] = None,
|
|
491
523
|
trace_id: str = "",
|
|
492
524
|
store: Optional[ContractRunStore] = None,
|
|
@@ -500,6 +532,10 @@ async def execute_turn(
|
|
|
500
532
|
Returns the :class:`ExecutionResult`, or ``None`` when the run was not
|
|
501
533
|
claimable (idempotent replay, already running/terminal, cancelled, or
|
|
502
534
|
awaiting HITL) — in which case **no business action is taken**.
|
|
535
|
+
|
|
536
|
+
``system_prompt``/``history``/``mode`` carry the ADR-0005 D1 conversational
|
|
537
|
+
inputs to the provider seam. Omitting all three reproduces the
|
|
538
|
+
pre-extension single-shot turn exactly.
|
|
503
539
|
"""
|
|
504
540
|
store = store or get_contract_run_store()
|
|
505
541
|
llm = llm or get_llm_callable()
|
|
@@ -561,6 +597,9 @@ async def execute_turn(
|
|
|
561
597
|
model=resolved_model or "",
|
|
562
598
|
message=message,
|
|
563
599
|
parameters=parameters or _generation_parameters(evidence),
|
|
600
|
+
system_prompt=system_prompt,
|
|
601
|
+
history=history,
|
|
602
|
+
mode=mode,
|
|
564
603
|
)
|
|
565
604
|
except Exception as exc: # noqa: BLE001 — one terminal outcome (AC-052)
|
|
566
605
|
# US-014 — preserve the source category/retryability through the
|
{plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/agent_run/registry.py
RENAMED
|
@@ -264,9 +264,24 @@ class ContractRunStore:
|
|
|
264
264
|
|
|
265
265
|
def attach_execution_result(self, run_id: str, result: object) -> None:
|
|
266
266
|
"""Store the run-scoped execution result (model identities, usage,
|
|
267
|
-
outcome) for the E6 trace handoff
|
|
267
|
+
outcome) for the E6 trace handoff — internal, not the ``/v1`` contract.
|
|
268
|
+
|
|
269
|
+
Also projects the result's ``output_text`` onto the stored resource so
|
|
270
|
+
every read path (``get``/``cancel``/``mark_terminal``) surfaces the reply
|
|
271
|
+
consistently (ADR-0005 D1). Both terminal helpers — ``_fail`` and
|
|
272
|
+
``_finalize`` — call this, so stamping here covers success, failure, and
|
|
273
|
+
cancellation from one site. Model identities and token usage stay
|
|
274
|
+
internal: only the reply text is contractual.
|
|
275
|
+
"""
|
|
276
|
+
text = getattr(result, "output_text", None)
|
|
268
277
|
with self._lock:
|
|
269
278
|
self._execution[run_id] = result
|
|
279
|
+
if text is not None:
|
|
280
|
+
resource = self._runs.get(run_id)
|
|
281
|
+
if resource is not None:
|
|
282
|
+
self._runs[run_id] = resource.model_copy(
|
|
283
|
+
update={"output_text": text}
|
|
284
|
+
)
|
|
270
285
|
|
|
271
286
|
def get_execution_result(self, run_id: str) -> Optional[object]:
|
|
272
287
|
with self._lock:
|
|
@@ -59,6 +59,7 @@ from __future__ import annotations
|
|
|
59
59
|
import argparse
|
|
60
60
|
import logging
|
|
61
61
|
import os
|
|
62
|
+
import uuid
|
|
62
63
|
from pathlib import Path
|
|
63
64
|
from typing import List, Optional
|
|
64
65
|
|
|
@@ -67,6 +68,7 @@ from fastapi import FastAPI
|
|
|
67
68
|
from plm_shared.cors import PlmCorsSettings, install_cors
|
|
68
69
|
from plm_shared.middleware import PlmTenantMiddleware
|
|
69
70
|
from plm_shared.middleware.traceparent import TraceparentMiddleware
|
|
71
|
+
from plm_shared.trace_context import new_root_context, reset_context, set_context
|
|
70
72
|
|
|
71
73
|
from . import __version__
|
|
72
74
|
from .routers import agent_runs as agent_runs_router
|
|
@@ -78,6 +80,38 @@ from .settings import EngineHostSettings, _resolve_install_identity
|
|
|
78
80
|
logger = logging.getLogger(__name__)
|
|
79
81
|
|
|
80
82
|
|
|
83
|
+
class DevTenantMiddleware:
|
|
84
|
+
"""DEV-ONLY ASGI middleware: bind a fixed tenant when the real tenant
|
|
85
|
+
middleware is disabled.
|
|
86
|
+
|
|
87
|
+
In production ``PlmTenantMiddleware`` (frozen) derives the tenant from a
|
|
88
|
+
validated ``X-Core-Caller`` header and 400s requests without it. In the
|
|
89
|
+
dev/SPA-parity launch that middleware is off and the web-shell calls
|
|
90
|
+
``:8000`` (config/models, health, process generation) with no header — so
|
|
91
|
+
those calls would 400, and governed ``/v1/agent-runs`` turns would 401 for
|
|
92
|
+
lack of a bound tenant. This shim mirrors ``PlmTenantMiddleware``'s context
|
|
93
|
+
binding with a fixed configured dev tenant so an unheadered dev request is
|
|
94
|
+
admitted with a resolved tenant. Installed ONLY when the real tenant
|
|
95
|
+
middleware is off AND ``ENGINE_HOST_DEV_TENANT`` is set; production never
|
|
96
|
+
installs it. Plain ASGI (mirrors the frozen middleware, not
|
|
97
|
+
``BaseHTTPMiddleware``).
|
|
98
|
+
"""
|
|
99
|
+
|
|
100
|
+
def __init__(self, app, tenant_id: str):
|
|
101
|
+
self._app = app
|
|
102
|
+
self._tenant_id = str(uuid.UUID(str(tenant_id)))
|
|
103
|
+
|
|
104
|
+
async def __call__(self, scope, receive, send) -> None:
|
|
105
|
+
if scope.get("type") != "http":
|
|
106
|
+
await self._app(scope, receive, send)
|
|
107
|
+
return
|
|
108
|
+
handle = set_context(new_root_context(tenant_id=self._tenant_id))
|
|
109
|
+
try:
|
|
110
|
+
await self._app(scope, receive, send)
|
|
111
|
+
finally:
|
|
112
|
+
reset_context(handle)
|
|
113
|
+
|
|
114
|
+
|
|
81
115
|
def _wire_identity(app: FastAPI) -> bool:
|
|
82
116
|
"""Install IdentityMiddleware + Hs256JwtIdentityProvider.
|
|
83
117
|
|
|
@@ -362,6 +396,19 @@ def build_app(
|
|
|
362
396
|
"ENGINE_HOST_INSTALL_TENANT_MIDDLEWARE=0 (dev parity with "
|
|
363
397
|
"legacy backend pre-ENABLE_PLM_IDENTITY_MIDDLEWARE wiring)"
|
|
364
398
|
)
|
|
399
|
+
# DEV-ONLY: with the real tenant middleware off, optionally bind a
|
|
400
|
+
# fixed dev tenant so unheadered SPA calls to :8000 aren't 400'd and
|
|
401
|
+
# governed /v1/agent-runs turns resolve a tenant. Never in production.
|
|
402
|
+
if host_settings.engine_host_dev_tenant:
|
|
403
|
+
# Fail loud at construction (middleware is instantiated lazily).
|
|
404
|
+
dev_tenant = str(uuid.UUID(str(host_settings.engine_host_dev_tenant)))
|
|
405
|
+
app.add_middleware(DevTenantMiddleware, tenant_id=dev_tenant)
|
|
406
|
+
logger.warning(
|
|
407
|
+
"engine_host_dev_tenant: DEV tenant binding active (tenant=%s) "
|
|
408
|
+
"— NOT for production; requests are bound to this tenant with "
|
|
409
|
+
"no X-Core-Caller. Set only in local/dev launches.",
|
|
410
|
+
host_settings.engine_host_dev_tenant,
|
|
411
|
+
)
|
|
365
412
|
if install_identity:
|
|
366
413
|
_wire_identity(app)
|
|
367
414
|
install_cors(app, settings=cors_settings)
|
{plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/core_host/routers/agent_runs.py
RENAMED
|
@@ -179,6 +179,16 @@ async def create_agent_run(
|
|
|
179
179
|
autonomy_level=autonomy_level,
|
|
180
180
|
parameters=body.agent.model.parameters.model_dump(exclude_none=True),
|
|
181
181
|
skills=body.agent.skills,
|
|
182
|
+
# ADR-0005 D1 — conversational inputs. `system_prompt` rides the
|
|
183
|
+
# snapshot (agent configuration, digest-covered); `history`/`mode`
|
|
184
|
+
# are turn-scoped and ride the request body.
|
|
185
|
+
system_prompt=body.agent.system_prompt,
|
|
186
|
+
history=(
|
|
187
|
+
[m.model_dump(mode="json") for m in body.history]
|
|
188
|
+
if body.history
|
|
189
|
+
else None
|
|
190
|
+
),
|
|
191
|
+
mode=body.mode.value if body.mode else None,
|
|
182
192
|
tenant_id=str(tenant_id),
|
|
183
193
|
trace_id=resource.trace_id,
|
|
184
194
|
)
|
|
@@ -89,6 +89,16 @@ class EngineHostSettings(BaseSettings):
|
|
|
89
89
|
# boot preserved.
|
|
90
90
|
engine_host_mcrc_emitter_tenant: Optional[str] = None
|
|
91
91
|
|
|
92
|
+
# DEV-ONLY tenant binding. When the real tenant middleware is disabled
|
|
93
|
+
# (``engine_host_install_tenant_middleware=False`` — the dev/SPA-parity
|
|
94
|
+
# launch where the web-shell calls :8000 with no ``X-Core-Caller``), a set
|
|
95
|
+
# value binds this tenant per request so (a) unheadered SPA calls are not
|
|
96
|
+
# 400'd by the rejecting middleware and (b) ``current_principal()`` finds a
|
|
97
|
+
# tenant, admitting governed ``/v1/agent-runs`` turns. IGNORED when the real
|
|
98
|
+
# tenant middleware is installed (production). Never enable in delivery.
|
|
99
|
+
# Set by ``run.py`` full mode alongside ``engine_host_mcrc_emitter_tenant``.
|
|
100
|
+
engine_host_dev_tenant: Optional[str] = None
|
|
101
|
+
|
|
92
102
|
class Config:
|
|
93
103
|
env_prefix = ""
|
|
94
104
|
case_sensitive = False
|
{plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/agent_telemetry_service.py
RENAMED
|
@@ -22,7 +22,22 @@ Import-rewires applied at lift:
|
|
|
22
22
|
|
|
23
23
|
Fire-and-forget persistence of LLM invocation records. Never blocks
|
|
24
24
|
callers. Failures are logged, never propagated.
|
|
25
|
+
|
|
26
|
+
**Provider-import hygiene (ADR-0005 D4).** ``LlmUsage`` is used here in
|
|
27
|
+
*type positions only* (the :class:`InvocationTracker` field and the
|
|
28
|
+
``record_invocation`` signature). Importing it eagerly pulled the whole
|
|
29
|
+
``llm_service`` → LiteLLM chain into every process that imports this
|
|
30
|
+
telemetry module — which is what made `litellm` load in Studio, whose
|
|
31
|
+
only use of this module is ``record_invocation`` /
|
|
32
|
+
``InvocationTracker`` / ``stamp_cancelled``. The import is therefore
|
|
33
|
+
``TYPE_CHECKING``-only, and PEP 563 (``from __future__ import
|
|
34
|
+
annotations``) makes the annotations lazy strings so nothing resolves
|
|
35
|
+
``LlmUsage`` at runtime. Safe here because this module defines only a
|
|
36
|
+
plain class and a plain ``@dataclass`` — no Pydantic/SQLModel model and
|
|
37
|
+
no ``get_type_hints()`` call resolves annotations eagerly.
|
|
25
38
|
"""
|
|
39
|
+
from __future__ import annotations
|
|
40
|
+
|
|
26
41
|
import asyncio
|
|
27
42
|
import json
|
|
28
43
|
import logging
|
|
@@ -32,11 +47,12 @@ import uuid
|
|
|
32
47
|
from concurrent.futures import ThreadPoolExecutor
|
|
33
48
|
from dataclasses import dataclass, field
|
|
34
49
|
from datetime import datetime, timezone
|
|
35
|
-
from typing import Optional
|
|
50
|
+
from typing import TYPE_CHECKING, Optional
|
|
36
51
|
|
|
37
52
|
from sqlmodel import Session
|
|
38
53
|
|
|
39
|
-
|
|
54
|
+
if TYPE_CHECKING: # pragma: no cover - typing-only, see module docstring
|
|
55
|
+
from plm_engine_core.runtime.llm_service import LlmUsage
|
|
40
56
|
|
|
41
57
|
|
|
42
58
|
class _SettingsResolver:
|
{plm_engine_core-1.3.0 → plm_engine_core-1.4.0}/plm_engine_core/runtime/invocation_telemetry.py
RENAMED
|
@@ -24,12 +24,19 @@ move + an opt-in `stamp_invocation` entry point. Call-site migration in
|
|
|
24
24
|
from __future__ import annotations
|
|
25
25
|
|
|
26
26
|
import logging
|
|
27
|
-
from typing import Optional
|
|
27
|
+
from typing import TYPE_CHECKING, Optional
|
|
28
28
|
|
|
29
29
|
from plm_engine_core.runtime.invocation_context import InvocationContext
|
|
30
30
|
from plm_shared.invocation_kinds import coerce_kind
|
|
31
31
|
from plm_shared.system_context import coerce_agent_id, coerce_workspace_id
|
|
32
|
-
|
|
32
|
+
|
|
33
|
+
if TYPE_CHECKING: # pragma: no cover - typing-only (ADR-0005 D4)
|
|
34
|
+
# ``LlmUsage`` is referenced only in a signature annotation, and this module
|
|
35
|
+
# already has PEP 563, so a lazy import keeps the LiteLLM chain out of every
|
|
36
|
+
# process that imports telemetry. This is the second link of the chain
|
|
37
|
+
# `agent_telemetry_service -> invocation_telemetry -> llm_service -> litellm`;
|
|
38
|
+
# guarding only the first link left `litellm` loading.
|
|
39
|
+
from plm_engine_core.runtime.llm_service import LlmUsage
|
|
33
40
|
|
|
34
41
|
logger = logging.getLogger(__name__)
|
|
35
42
|
|
|
@@ -498,13 +498,21 @@ Return ONLY the raw JSON object.
|
|
|
498
498
|
temperature: float = 0.7,
|
|
499
499
|
model: str | None = None,
|
|
500
500
|
max_tokens: int | None = None,
|
|
501
|
+
history: list[dict] | None = None,
|
|
501
502
|
) -> Tuple[str, LlmUsage]:
|
|
502
|
-
"""Like generate_reply but also returns token usage metadata.
|
|
503
|
+
"""Like generate_reply but also returns token usage metadata.
|
|
504
|
+
|
|
505
|
+
``history`` (ADR-0005 D1) replays prior role-tagged turns between the
|
|
506
|
+
system and the current user message, so a governed multi-turn run keeps
|
|
507
|
+
real roles instead of a flattened transcript. Trailing keyword with a
|
|
508
|
+
default — every existing positional caller is unaffected.
|
|
509
|
+
"""
|
|
503
510
|
resolved_model = model or self.model
|
|
504
511
|
kwargs: dict = dict(
|
|
505
512
|
model=resolved_model,
|
|
506
513
|
messages=[
|
|
507
514
|
{"role": "system", "content": system_prompt},
|
|
515
|
+
*(list(history) if history else []),
|
|
508
516
|
{"role": "user", "content": user_prompt},
|
|
509
517
|
],
|
|
510
518
|
temperature=temperature,
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: plm-engine-core
|
|
3
|
-
Version: 1.
|
|
3
|
+
Version: 1.4.0
|
|
4
4
|
Summary: TracePulse PLM Engine Core — control_plane (governance, autonomy, HITL, run lifecycle, trace propagation) + agent_runtime (selector, dispatcher, retry, escalation). US-CR.1 scaffold; behaviour lands in US-CR.0+. Internal one-way import contract: control_plane MUST NOT import from agent_runtime (mechanical enforcement in US-CR.2).
|
|
5
5
|
Author: TracePulse
|
|
6
6
|
License: Proprietary
|
|
@@ -112,6 +112,7 @@ tests/test_contract_bundle.py
|
|
|
112
112
|
tests/test_core_host_build_app.py
|
|
113
113
|
tests/test_core_host_wave9_parity_smoke.py
|
|
114
114
|
tests/test_cost_budget_policy.py
|
|
115
|
+
tests/test_dev_tenant_middleware.py
|
|
115
116
|
tests/test_error_code_mirror.py
|
|
116
117
|
tests/test_error_taxonomy.py
|
|
117
118
|
tests/test_evidence_pack.py
|
|
@@ -120,6 +121,7 @@ tests/test_execution_error_normalization.py
|
|
|
120
121
|
tests/test_g8_conformance.py
|
|
121
122
|
tests/test_gate_verdict.py
|
|
122
123
|
tests/test_governance_chain.py
|
|
124
|
+
tests/test_governed_turn_contract_extension.py
|
|
123
125
|
tests/test_health_readiness.py
|
|
124
126
|
tests/test_hitl_gate.py
|
|
125
127
|
tests/test_hitl_resume.py
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "plm-engine-core"
|
|
7
|
-
version = "1.
|
|
7
|
+
version = "1.4.0"
|
|
8
8
|
description = "TracePulse PLM Engine Core — control_plane (governance, autonomy, HITL, run lifecycle, trace propagation) + agent_runtime (selector, dispatcher, retry, escalation). US-CR.1 scaffold; behaviour lands in US-CR.0+. Internal one-way import contract: control_plane MUST NOT import from agent_runtime (mechanical enforcement in US-CR.2)."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.12"
|
|
@@ -73,7 +73,7 @@ class _FakeLlm:
|
|
|
73
73
|
)
|
|
74
74
|
self._raises = raises
|
|
75
75
|
|
|
76
|
-
async def __call__(self, *, model, message, parameters):
|
|
76
|
+
async def __call__(self, *, model, message, parameters, **_):
|
|
77
77
|
self.calls.append({"model": model, "message": message, "parameters": dict(parameters)})
|
|
78
78
|
if self._raises is not None:
|
|
79
79
|
raise self._raises
|
|
@@ -184,7 +184,7 @@ async def test_ac052_cancel_after_provider_boundary_wins() -> None:
|
|
|
184
184
|
store = get_contract_run_store()
|
|
185
185
|
run_id = _queued_run(store)
|
|
186
186
|
|
|
187
|
-
async def cancelling_llm(*, model, message, parameters):
|
|
187
|
+
async def cancelling_llm(*, model, message, parameters, **_):
|
|
188
188
|
store.cancel(run_id) # cooperative cancel lands during the turn
|
|
189
189
|
return LlmTurnResult(text="late", used_model="openai/gpt-4o")
|
|
190
190
|
|
|
@@ -0,0 +1,110 @@
|
|
|
1
|
+
"""DevTenantMiddleware (core_host) — dev-only tenant binding for middleware-off.
|
|
2
|
+
|
|
3
|
+
In production ``PlmTenantMiddleware`` (frozen) 400s any :8000 request without a
|
|
4
|
+
valid ``X-Core-Caller``. In the dev/SPA-parity launch that middleware is off and
|
|
5
|
+
the web-shell calls :8000 with no header, so those calls would 400 and governed
|
|
6
|
+
``/v1/agent-runs`` turns would 401 for lack of a bound tenant. ``ENGINE_HOST_DEV_TENANT``
|
|
7
|
+
installs ``DevTenantMiddleware`` in that (and only that) case, binding a fixed
|
|
8
|
+
tenant per request.
|
|
9
|
+
|
|
10
|
+
Proven here:
|
|
11
|
+
* the middleware binds the configured tenant into ``current_context()`` for an
|
|
12
|
+
HTTP request and resets it afterwards (no leak); non-HTTP scopes pass through;
|
|
13
|
+
* build_app installs it ONLY when the real tenant middleware is off AND
|
|
14
|
+
``engine_host_dev_tenant`` is set — inert in production;
|
|
15
|
+
* an unheadered request is admitted (200, not the middleware 400) when bound;
|
|
16
|
+
* a malformed dev-tenant UUID fails loud at construction.
|
|
17
|
+
"""
|
|
18
|
+
from __future__ import annotations
|
|
19
|
+
|
|
20
|
+
import uuid
|
|
21
|
+
|
|
22
|
+
import pytest
|
|
23
|
+
from fastapi import FastAPI
|
|
24
|
+
from httpx import ASGITransport, AsyncClient
|
|
25
|
+
|
|
26
|
+
from plm_shared.trace_context import current_context
|
|
27
|
+
from plm_engine_core.core_host.main import DevTenantMiddleware, build_app
|
|
28
|
+
from plm_engine_core.core_host.settings import EngineHostSettings
|
|
29
|
+
|
|
30
|
+
_DEV_TENANT = "e0b41f0c-c55c-43e8-b534-e375ea8b4ba2"
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
def _mw_names(app: FastAPI) -> list:
|
|
34
|
+
return [m.cls.__name__ for m in app.user_middleware]
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
@pytest.mark.asyncio
|
|
38
|
+
async def test_middleware_binds_and_resets_context() -> None:
|
|
39
|
+
captured = {}
|
|
40
|
+
|
|
41
|
+
async def _inner(scope, receive, send):
|
|
42
|
+
ctx = current_context()
|
|
43
|
+
captured["tenant"] = ctx.tenant_id if ctx else None
|
|
44
|
+
|
|
45
|
+
mw = DevTenantMiddleware(_inner, tenant_id=_DEV_TENANT)
|
|
46
|
+
await mw({"type": "http", "headers": []}, None, None) # no X-Core-Caller
|
|
47
|
+
|
|
48
|
+
assert captured["tenant"] == _DEV_TENANT # bound for the request
|
|
49
|
+
assert current_context() is None # reset afterwards (no leak)
|
|
50
|
+
|
|
51
|
+
|
|
52
|
+
@pytest.mark.asyncio
|
|
53
|
+
async def test_non_http_scope_passes_through() -> None:
|
|
54
|
+
seen = {}
|
|
55
|
+
|
|
56
|
+
async def _inner(scope, receive, send):
|
|
57
|
+
seen["type"] = scope["type"]
|
|
58
|
+
seen["ctx"] = current_context()
|
|
59
|
+
|
|
60
|
+
mw = DevTenantMiddleware(_inner, tenant_id=_DEV_TENANT)
|
|
61
|
+
await mw({"type": "lifespan"}, None, None)
|
|
62
|
+
assert seen["type"] == "lifespan"
|
|
63
|
+
assert seen["ctx"] is None # not bound for non-http
|
|
64
|
+
|
|
65
|
+
|
|
66
|
+
def test_build_app_installs_dev_when_tenant_middleware_off() -> None:
|
|
67
|
+
settings = EngineHostSettings(
|
|
68
|
+
engine_host_install_tenant_middleware=False,
|
|
69
|
+
engine_host_dev_tenant=_DEV_TENANT,
|
|
70
|
+
)
|
|
71
|
+
app = build_app(install_identity=False, skip_startup=True, host_settings=settings)
|
|
72
|
+
names = _mw_names(app)
|
|
73
|
+
assert "DevTenantMiddleware" in names
|
|
74
|
+
assert "PlmTenantMiddleware" not in names # real middleware stays off
|
|
75
|
+
|
|
76
|
+
|
|
77
|
+
def test_build_app_no_dev_when_tenant_middleware_on() -> None:
|
|
78
|
+
# Real middleware ON + a dev tenant set → the dev shim is NEVER installed.
|
|
79
|
+
settings = EngineHostSettings(
|
|
80
|
+
engine_host_install_tenant_middleware=True,
|
|
81
|
+
engine_host_dev_tenant=_DEV_TENANT,
|
|
82
|
+
)
|
|
83
|
+
app = build_app(install_identity=False, skip_startup=True, host_settings=settings)
|
|
84
|
+
names = _mw_names(app)
|
|
85
|
+
assert "PlmTenantMiddleware" in names
|
|
86
|
+
assert "DevTenantMiddleware" not in names
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
@pytest.mark.asyncio
|
|
90
|
+
async def test_unheadered_request_admitted_when_dev_bound() -> None:
|
|
91
|
+
settings = EngineHostSettings(
|
|
92
|
+
engine_host_install_tenant_middleware=False,
|
|
93
|
+
engine_host_dev_tenant=_DEV_TENANT,
|
|
94
|
+
)
|
|
95
|
+
app = build_app(install_identity=False, skip_startup=True, host_settings=settings)
|
|
96
|
+
transport = ASGITransport(app=app)
|
|
97
|
+
async with AsyncClient(transport=transport, base_url="http://test") as client:
|
|
98
|
+
# No X-Core-Caller — with the real middleware this 400s; the dev
|
|
99
|
+
# binding admits it (the SPA-unheadered-call fix).
|
|
100
|
+
resp = await client.get("/v1/health")
|
|
101
|
+
assert resp.status_code == 200
|
|
102
|
+
|
|
103
|
+
|
|
104
|
+
def test_build_app_malformed_dev_tenant_fails_loud() -> None:
|
|
105
|
+
settings = EngineHostSettings(
|
|
106
|
+
engine_host_install_tenant_middleware=False,
|
|
107
|
+
engine_host_dev_tenant="not-a-uuid",
|
|
108
|
+
)
|
|
109
|
+
with pytest.raises(ValueError):
|
|
110
|
+
build_app(install_identity=False, skip_startup=True, host_settings=settings)
|
|
@@ -140,7 +140,7 @@ async def test_ac060_executor_maps_content_filter_end_to_end() -> None:
|
|
|
140
140
|
store = get_contract_run_store()
|
|
141
141
|
run_id = _queued_run(store)
|
|
142
142
|
|
|
143
|
-
async def filtered_llm(*, model, message, parameters):
|
|
143
|
+
async def filtered_llm(*, model, message, parameters, **_):
|
|
144
144
|
raise _model_error(ModelErrorCode.MODEL_CONTENT_FILTERED)
|
|
145
145
|
|
|
146
146
|
result = await execute_turn(run_id, message="m", store=store, llm=filtered_llm)
|
|
@@ -169,7 +169,7 @@ async def test_ac061_terminal_signal_emitted_exactly_once_on_failure() -> None:
|
|
|
169
169
|
run_id = _queued_run(store)
|
|
170
170
|
sink = _CountingSink()
|
|
171
171
|
|
|
172
|
-
async def boom(*, model, message, parameters):
|
|
172
|
+
async def boom(*, model, message, parameters, **_):
|
|
173
173
|
raise _model_error(ModelErrorCode.MODEL_PROVIDER_ERROR)
|
|
174
174
|
|
|
175
175
|
await execute_turn(run_id, message="m", store=store, llm=boom, sink=sink)
|
|
@@ -184,7 +184,7 @@ async def test_ac061_terminal_signal_emitted_exactly_once_on_success() -> None:
|
|
|
184
184
|
run_id = _queued_run(store)
|
|
185
185
|
sink = _CountingSink()
|
|
186
186
|
|
|
187
|
-
async def ok_llm(*, model, message, parameters):
|
|
187
|
+
async def ok_llm(*, model, message, parameters, **_):
|
|
188
188
|
return LlmTurnResult(text="ok", used_model="openai/gpt-4o")
|
|
189
189
|
|
|
190
190
|
await execute_turn(run_id, message="m", store=store, llm=ok_llm, sink=sink)
|