MCP 0.4.4: pin hosted images to Chaos 0.9.0

CHANGELOG.md

@@ -6,6 +6,22 @@ This project follows [Semantic Versioning](https://semver.org/).
 
 ## [Unreleased]
 
+## 0.4.4 - 2026-10-31
+
+### Added
+
+- Agentic Chaos 0.9.0 (roadmap v0.9 Resilience Probes & Resilience Score)
+  compatibility: `chaos.export_aios_run` passes a scored report's derived
+  `resilience_score` measurement and its probe/score producer extension
+  through unchanged (`tests/test_chaos_aios_export.py`). Probes, scoring and
+  run history stay Python/CLI capabilities; no new MCP tool, and hosted HTTP
+  still rejects `chaos.run_experiment`. See `docs/chaos-0.9.md`.
+
+### Changed
+
+- Hosted images pin `agentic-chaos==0.9.0` and `agentic-sidecar==0.9.1`;
+  `tests/test_hosted_image.py` requires Chaos 0.9.0 or later.
+
 ## 0.4.3 - 2026-10-31
 
 ### Added

README.md

@@ -136,7 +136,7 @@ and `chaos.export_aios_run` behave the same hosted as locally. Only
 
 Both need `agenticlens>=0.7.0`; with an older install they return a clear
 "unavailable" error and every other Lens tool keeps working. The hosted images
-pin `agenticlens==0.11.0` (since MCP 0.4.3) and `agentic-chaos==0.8.0` (since MCP 0.3.13).
+pin `agenticlens==0.11.0` (since MCP 0.4.3) and `agentic-chaos==0.9.0` (since MCP 0.4.4).
 
 ## AgenticLens 0.8.0 suite tools (MCP 0.3.9)
 
@@ -697,3 +697,38 @@ See [operator compatibility and ownership](docs/tower-0.8-live-operations.md) fo
 periodic detection, durable alerts, acknowledgements and joined incident console
 views. Existing native contracts and optional dependency ranges remain compatible;
 no runtime or patch release is required here.
+
+## Chaos 0.9.0 interoperability
+
+Chaos 0.9.0 adds:
+
+- resilience probes: `HttpProbe`, `ResponseQualityProbe`, `LatencyProbe` and
+  `FunctionProbe` check steady state before chaos, while faults are active and
+  after recovery. By default no fault is injected when a pre-chaos probe
+  fails; `require_steady_state=False` runs the chaos anyway and leaves the
+  score inconclusive
+- a 0-100 resilience score built from recovery, probe, latency,
+  retry-efficiency and cost evidence, with per-fault verdicts. Components a
+  run cannot show (for example latency without a baseline) are left out and
+  the remaining weights renormalized. The score is inconclusive (`null`) after
+  a failed pre-chaos probe, for dry runs, when no fault fired, or when no
+  component can be computed
+- a local run-history store for score trends, with `max_runs` and
+  `max_age_days` retention, and a `--min-score` CI gate
+- probes and scoring on chaos workflows, experiment files and the pytest marker
+
+Native event IDs and fault types are unchanged. `ChaosReport.probes`,
+`ChaosReport.resilience` and the matching `WorkflowReport` fields are optional
+and additive. When a score was computed, AIOS exports add a derived
+`resilience_score` measurement; inconclusive runs export none. Probe results
+and the score breakdown stay in the `io.deepagentlabs.agentic-chaos`
+producer extension.
+
+See the [v0.9 guide](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/docs/v0.9.md)
+and the [offline demo](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/examples/chaos_resilience_demo.py).
+The resilience score summarizes the evidence of one chaos run. It is not an
+Evals score, a release-gate decision, a deployment health signal or proof of
+containment. No new shared schema, automatic adapter or hosted execution
+capability is introduced.
+
+MCP 0.4.4 pins the published images to Chaos 0.9.0. `chaos.export_aios_run` passes the `resilience_score` measurement and the producer extension through unchanged, and the existing registry discovers faults. Probes, scoring and history are Python/CLI capabilities, not new MCP tools. Hosted HTTP continues to reject `chaos.run_experiment`, so the hosted service neither executes experiment targets nor calls probe endpoints.

ROADMAP.md

@@ -872,3 +872,38 @@ and incident console views. Existing local authenticated tower.* investigation/r
 
 See [compatibility and operator example](docs/tower-0.8-live-operations.md).
 No sibling runtime change or patch release is required.
+
+## Chaos 0.9.0 interoperability
+
+Chaos 0.9.0 adds:
+
+- resilience probes: `HttpProbe`, `ResponseQualityProbe`, `LatencyProbe` and
+  `FunctionProbe` check steady state before chaos, while faults are active and
+  after recovery. By default no fault is injected when a pre-chaos probe
+  fails; `require_steady_state=False` runs the chaos anyway and leaves the
+  score inconclusive
+- a 0-100 resilience score built from recovery, probe, latency,
+  retry-efficiency and cost evidence, with per-fault verdicts. Components a
+  run cannot show (for example latency without a baseline) are left out and
+  the remaining weights renormalized. The score is inconclusive (`null`) after
+  a failed pre-chaos probe, for dry runs, when no fault fired, or when no
+  component can be computed
+- a local run-history store for score trends, with `max_runs` and
+  `max_age_days` retention, and a `--min-score` CI gate
+- probes and scoring on chaos workflows, experiment files and the pytest marker
+
+Native event IDs and fault types are unchanged. `ChaosReport.probes`,
+`ChaosReport.resilience` and the matching `WorkflowReport` fields are optional
+and additive. When a score was computed, AIOS exports add a derived
+`resilience_score` measurement; inconclusive runs export none. Probe results
+and the score breakdown stay in the `io.deepagentlabs.agentic-chaos`
+producer extension.
+
+See the [v0.9 guide](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/docs/v0.9.md)
+and the [offline demo](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/examples/chaos_resilience_demo.py).
+The resilience score summarizes the evidence of one chaos run. It is not an
+Evals score, a release-gate decision, a deployment health signal or proof of
+containment. No new shared schema, automatic adapter or hosted execution
+capability is introduced.
+
+MCP 0.4.4 pins the published images to Chaos 0.9.0. `chaos.export_aios_run` passes the `resilience_score` measurement and the producer extension through unchanged, and the existing registry discovers faults. Probes, scoring and history are Python/CLI capabilities, not new MCP tools. Hosted HTTP continues to reject `chaos.run_experiment`, so the hosted service neither executes experiment targets nor calls probe endpoints.

deploy/docker/Dockerfile.aws

@@ -3,13 +3,15 @@ FROM python:3.12-slim
 WORKDIR /app
 COPY pyproject.toml README.md ./
 COPY src ./src
-# Pairs: Lens 0.10.0 (diagnosis, suites, ModelFit, advisory routing),
-# Chaos 0.8.0 (AIOS run export; extension carries call counts and workflow summaries),
+# Pairs: Lens 0.11.0 (diagnosis, suites, ModelFit, advisory routing, optimization analysis),
+# Chaos 0.9.0 (AIOS run export; extension carries call counts, workflow summaries, probe
+# results and the resilience score, plus a derived resilience_score measurement),
+# Sidecar 0.9.1,
 # Evals 0.10.0 (the producer of the native report/gate schemas bundled for evals.validate_artifact;
 # unchanged since 0.9.0) and of AIOS evaluation evidence that spec.validate_artifact checks.
 # Chaos is installed for its exporter and fault catalog only; HTTP always
 # rejects chaos.run_experiment, so no target script is ever executed here.
-RUN pip install --no-cache-dir ".[http]" "agenticlens==0.11.0" "agentic-chaos==0.8.0" "agentic-evals==0.10.0" "agentic-sidecar==0.9.0"
+RUN pip install --no-cache-dir ".[http]" "agenticlens==0.11.0" "agentic-chaos==0.9.0" "agentic-evals==0.10.0" "agentic-sidecar==0.9.1"
 COPY scripts/fetch_schema_assets.py /tmp/fetch_schema_assets.py
 RUN python /tmp/fetch_schema_assets.py /opt/workspace/ai-operations-spec
 RUN useradd --create-home --uid 10001 mcp

deploy/docker/Dockerfile.render

@@ -3,13 +3,15 @@ FROM python:3.12-slim
 WORKDIR /app
 COPY pyproject.toml README.md ./
 COPY src ./src
-# Pairs: Lens 0.10.0 (diagnosis, suites, ModelFit, advisory routing),
-# Chaos 0.8.0 (AIOS run export; extension carries call counts and workflow summaries),
+# Pairs: Lens 0.11.0 (diagnosis, suites, ModelFit, advisory routing, optimization analysis),
+# Chaos 0.9.0 (AIOS run export; extension carries call counts, workflow summaries, probe
+# results and the resilience score, plus a derived resilience_score measurement),
+# Sidecar 0.9.1,
 # Evals 0.10.0 (the producer of the native report/gate schemas bundled for evals.validate_artifact;
 # unchanged since 0.9.0) and of AIOS evaluation evidence that spec.validate_artifact checks.
 # Chaos is installed for its exporter and fault catalog only; HTTP always
 # rejects chaos.run_experiment, so no target script is ever executed here.
-RUN pip install --no-cache-dir ".[http]" "agenticlens==0.11.0" "agentic-chaos==0.8.0" "agentic-evals==0.10.0" "agentic-sidecar==0.9.0"
+RUN pip install --no-cache-dir ".[http]" "agenticlens==0.11.0" "agentic-chaos==0.9.0" "agentic-evals==0.10.0" "agentic-sidecar==0.9.1"
 COPY scripts/fetch_schema_assets.py /tmp/fetch_schema_assets.py
 RUN python /tmp/fetch_schema_assets.py /opt/workspace/ai-operations-spec
 RUN useradd --create-home --uid 10001 mcp

deploy/docker/README.md

@@ -13,7 +13,7 @@ python3 scripts/build_container.py --target workspace --tag deep-agentic-core-mc
 
 `workspace` is the default and needs sibling `agenticlens`, `agentic-chaos`
 and `ai-operations-spec` checkouts. AWS and Render builds use this repository's
-source, published `agenticlens==0.11.0` and `agentic-chaos==0.8.0`, and
+source, published `agenticlens==0.11.0` and `agentic-chaos==0.9.0`, and
 checksum-verified `ai-operations-spec` v0.5.1 assets (schemas plus the
 conformance and compatibility catalogs) fetched by `scripts/fetch_schema_assets.py`.
 Agentic Chaos is present only for its AIOS exporter and fault catalog; the HTTP

docs/chaos-0.9.md

@@ -0,0 +1,36 @@
+# Chaos 0.9 compatibility
+
+## Chaos 0.9.0 interoperability
+
+Chaos 0.9.0 adds:
+
+- resilience probes: `HttpProbe`, `ResponseQualityProbe`, `LatencyProbe` and
+  `FunctionProbe` check steady state before chaos, while faults are active and
+  after recovery. By default no fault is injected when a pre-chaos probe
+  fails; `require_steady_state=False` runs the chaos anyway and leaves the
+  score inconclusive
+- a 0-100 resilience score built from recovery, probe, latency,
+  retry-efficiency and cost evidence, with per-fault verdicts. Components a
+  run cannot show (for example latency without a baseline) are left out and
+  the remaining weights renormalized. The score is inconclusive (`null`) after
+  a failed pre-chaos probe, for dry runs, when no fault fired, or when no
+  component can be computed
+- a local run-history store for score trends, with `max_runs` and
+  `max_age_days` retention, and a `--min-score` CI gate
+- probes and scoring on chaos workflows, experiment files and the pytest marker
+
+Native event IDs and fault types are unchanged. `ChaosReport.probes`,
+`ChaosReport.resilience` and the matching `WorkflowReport` fields are optional
+and additive. When a score was computed, AIOS exports add a derived
+`resilience_score` measurement; inconclusive runs export none. Probe results
+and the score breakdown stay in the `io.deepagentlabs.agentic-chaos`
+producer extension.
+
+See the [v0.9 guide](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/docs/v0.9.md)
+and the [offline demo](https://github.com/DeepAgentLabs/agentic-chaos/blob/v0.9.0/examples/chaos_resilience_demo.py).
+The resilience score summarizes the evidence of one chaos run. It is not an
+Evals score, a release-gate decision, a deployment health signal or proof of
+containment. No new shared schema, automatic adapter or hosted execution
+capability is introduced.
+
+MCP 0.4.4 pins the published images to Chaos 0.9.0. `chaos.export_aios_run` passes the `resilience_score` measurement and the producer extension through unchanged, and the existing registry discovers faults. Probes, scoring and history are Python/CLI capabilities, not new MCP tools. Hosted HTTP continues to reject `chaos.run_experiment`, so the hosted service neither executes experiment targets nor calls probe endpoints.

mkdocs.yml

@@ -51,6 +51,7 @@ nav:
   - Unified Workflows: unified-workflows.md
   - Chaos 0.7 Compatibility: chaos-0.7.md
   - Chaos 0.8 Compatibility: chaos-0.8.md
+  - Chaos 0.9 Compatibility: chaos-0.9.md
   - Evals CLI Compatibility: evals-0.8-cli.md
   - Evals Report Contract: evals-report-contract.md
   - Evals 0.9 Judge Reliability: evals-0.9-judge-reliability.md

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "deep-agentic-core-mcp"
-version = "0.4.3"
+version = "0.4.4"
 description = "Unified MCP server for AgenticLens and Agentic Chaos workflows."
 readme = "README.md"
 requires-python = ">=3.10"

server.json

@@ -3,7 +3,7 @@
   "name": "io.github.DeepAgentLabs/deep-agentic-core-mcp",
   "title": "Deep Agentic Core MCP",
   "description": "Unified MCP server for AgenticLens and Agentic Chaos capabilities.",
-  "version": "0.4.3",
+  "version": "0.4.4",
   "repository": {
     "url": "https://github.com/DeepAgentLabs/mcp-server",
     "source": "github"
@@ -12,7 +12,7 @@
     {
       "registryType": "pypi",
       "identifier": "deep-agentic-core-mcp",
-      "version": "0.4.3",
+      "version": "0.4.4",
       "transport": {
         "type": "stdio"
       }

src/deep_agentic_core_mcp/__init__.py

@@ -2,4 +2,4 @@
 
 __all__ = ["__version__"]
 
-__version__ = "0.4.3"
+__version__ = "0.4.4"

tests/test_chaos_aios_export.py

@@ -175,3 +175,49 @@ async def test_export_tool_is_offered_over_remote_transport(
 
     assert "chaos.export_aios_run" in names
     assert "chaos.run_experiment" not in names
+
+
+def _chaos_version() -> tuple[int, ...]:
+    from importlib.metadata import PackageNotFoundError, version
+
+    try:
+        return tuple(int(part) for part in version("agentic-chaos").split(".")[:2])
+    except (PackageNotFoundError, ValueError):
+        return (0, 0)
+
+
+@pytest.mark.skipif(_chaos_version() < (0, 9), reason="resilience scores need agentic-chaos>=0.9")
+async def test_export_tool_passes_the_chaos_resilience_score_through() -> None:
+    report = {
+        "id": "report-9",
+        "name": "scored",
+        "start_time": "2026-10-31T12:00:00Z",
+        "end_time": "2026-10-31T12:00:01Z",
+        "chaos_events": [
+            {
+                "id": "e1",
+                "fault_type": "rate_limit_storm",
+                "step_name": "llm",
+                "outcome": "errored",
+                "message": "429",
+                "timestamp": "2026-10-31T12:00:00.5Z",
+            }
+        ],
+        "probes": [{"probe": "ping", "phase": "post", "status": "passed"}],
+        "resilience": {"version": "1.0", "score": 88, "status": "scored", "grade": "B"},
+    }
+
+    payload = await _call("chaos.export_aios_run", {"report": report, "session_id": "aios"})
+
+    assert payload["ok"] is True
+    run = payload["aios_run"]
+    measurements = {m["name"]: m for m in run["measurements"]}
+    assert measurements["resilience_score"] == {
+        "name": "resilience_score",
+        "value": 88,
+        "unit": "score",
+        "kind": "derived",
+    }
+    extension = run["extensions"]["io.deepagentlabs.agentic-chaos"]
+    assert extension["resilience"]["grade"] == "B"
+    assert extension["probes"][0]["probe"] == "ping"

tests/test_hosted_image.py

@@ -22,7 +22,7 @@
 # (agenticlens 0.7.0 adds lens.diagnose_run and lens.score_analysis; 0.8.0 adds
 # lens.suite_from_traces and lens.suite_check; 0.9.0 adds lens.modelfit_recommend;
 # 0.10.0 adds lens.routing_decide; 0.11.0 adds lens.optimization_analyze).
-REQUIRED_PINS = {"agenticlens": (0, 11, 0), "agentic-chaos": (0, 5, 1)}
+REQUIRED_PINS = {"agenticlens": (0, 11, 0), "agentic-chaos": (0, 9, 0)}
 
 # Spec-root-relative paths the ai_operations_spec adapter reads.
 SPEC_ASSETS = {
@@ -47,13 +47,13 @@ def test_standalone_images_pin_phase3_sibling_versions(dockerfile: str) -> None:
         assert _version(match.group(1)) >= minimum, f"{dockerfile} pins {package} too old"
 
 
-# Exact sibling releases the current hosted images ship (0.4.3: Lens 0.11.0, Chaos 0.8.0,
+# Exact sibling releases the current hosted images ship (0.4.4: Lens 0.11.0, Chaos 0.9.0,
 # Evals 0.10.0, whose native schemas are the 0.9.0 documents bundled here).
 HOSTED_PINS = {
     "agenticlens": "0.11.0",
-    "agentic-chaos": "0.8.0",
+    "agentic-chaos": "0.9.0",
     "agentic-evals": "0.10.0",
-    "agentic-sidecar": "0.9.0",
+    "agentic-sidecar": "0.9.1",
 }