Add agent harness integrations

.github/workflows/release.yml

@@ -44,8 +44,33 @@ jobs:
             'import site; print(site.getsitepackages()[0])')
           python -m pip_audit --path "$product_site"
 
+  harness-assurance:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - profile: langgraph
+            package: "langchain-core>=1,<2"
+          - profile: crewai
+            package: "crewai>=1,<2"
+          - profile: openai-agents
+            package: "openai-agents>=0.19,<1"
+    steps:
+      - uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5
+      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: "3.13"
+          cache: pip
+      - run: python -m pip install --upgrade pip
+      - run: python -m pip install -e '.[dev]' '${{ matrix.package }}'
+      - name: Exercise current public harness objects
+        env:
+          MENDMARK_HARNESS: ${{ matrix.profile }}
+        run: python -m pytest tests/test_harness_live_compatibility.py
+
   build:
-    needs: assurance
+    needs: [assurance, harness-assurance]
     runs-on: ubuntu-latest
     outputs:
       version: ${{ steps.package-version.outputs.version }}

.github/workflows/tests.yml

@@ -2,6 +2,7 @@ name: tests
 
 on:
   push:
+    branches: [main]
   pull_request:
   workflow_dispatch:
 
@@ -92,6 +93,31 @@ jobs:
         run: python -m pip install -e '.[deepeval,rubric,dev]'
       - run: python -m pytest tests/test_deepeval.py tests/test_rubric_example.py
 
+  harness-compatibility:
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - profile: langgraph
+            package: "langchain-core>=1,<2"
+          - profile: crewai
+            package: "crewai>=1,<2"
+          - profile: openai-agents
+            package: "openai-agents>=0.19,<1"
+    steps:
+      - uses: actions/checkout@fbc6f3992d24b796d5a048ff273f7fcc4a7b6c09 # v5
+      - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
+        with:
+          python-version: "3.13"
+          cache: pip
+      - run: python -m pip install --upgrade pip
+      - run: python -m pip install -e '.[dev]' '${{ matrix.package }}'
+      - name: Exercise current public harness objects
+        env:
+          MENDMARK_HARNESS: ${{ matrix.profile }}
+        run: python -m pytest tests/test_harness_live_compatibility.py
+
   distribution-assurance:
     runs-on: ubuntu-latest
     steps:

CHANGELOG.md

@@ -3,7 +3,24 @@
 All notable changes to Mendmark are documented here. The project follows
 Semantic Versioning for its Python and JSON contracts.
 
-## Unreleased
+## 0.6.0 - 2026-08-10
+
+### Added
+
+- Dependency-light, public-object adapters for LangChain/LangGraph messages,
+  CrewAI events, and OpenAI Agents SDK run items, including tool-schema and
+  side-effect metadata conversion.
+- `mendmark equip` for bounded harness detection, non-destructive local
+  scaffolding, an offline evaluator, an inactive pinned CI template, and a
+  copyable coding-agent self-equip prompt.
+- Live compatibility assurance against current releases of all three harness
+  paths, plus explicit human approval before observed traces can become
+  expected behavior.
+- A fluent causal-case builder for reviewed multi-agent delegation, parallel
+  dependencies, tool authority, state changes, results, and aggregation without
+  hand-authoring schema 2.0 JSON.
+
+## 0.5.0 - 2026-08-06
 
 ### Added
 

README.md

@@ -11,6 +11,7 @@
   </p>
   <p>
     <a href="#quick-start">Quick start</a> ·
+    <a href="#equip-an-agent-harness">Harnesses</a> ·
     <a href="#agent-eval-golden-set">Golden set</a> ·
     <a href="docs/multi-agent.md">Multi-agent</a> ·
     <a href="docs/agent-mutation-audits.md">How it works</a> ·
@@ -63,6 +64,31 @@ mendmark audit-json examples/multi_agent_suite.json \
   --evaluator-command "python3 examples/multi_agent_evaluator.py"
 ```
 
+## Equip an agent harness
+
+Mendmark has dependency-light adapters for LangChain/LangGraph, CrewAI, and the
+OpenAI Agents SDK. In an existing agent repository:
+
+```bash
+python -m pip install 'mendmark-evals==0.6.0'
+mendmark equip --framework auto
+```
+
+The command detects bounded dependency files and creates a reviewed capture
+guide, offline evaluator, and inactive CI template under `.mendmark/`. It does
+not edit application code, upload a trace, overwrite existing work, enable CI,
+or accept a baseline.
+
+Want the repository's coding agent to perform the integration?
+
+```bash
+mendmark equip --print-agent-prompt
+```
+
+See the [agent harness integration guide](docs/harness-integrations.md) for the
+direct Python APIs, explicit trace-approval boundary, framework compatibility,
+and multi-agent guidance.
+
 ## See the blind spot in two minutes
 
 **[▶ Watch the narrated weak-eval demonstration (original v1 fault inventory)](docs/assets/mendmark-weak-eval-demo.mp4)**
@@ -346,7 +372,7 @@ and the [ML evaluation card](https://github.com/danielgaskins/mendmark/blob/main
 
 ## Current boundary
 
-Version 0.5 is a local, open-source engine. It does not yet provide a hosted
+Version 0.6 is a local, open-source engine. It does not yet provide a hosted
 dashboard, team accounts, remote trace ingestion, or a secrets service. The
 planned control plane is described in [the product design](https://github.com/danielgaskins/mendmark/blob/main/docs/product.md).
 

docs/assurance.md

@@ -10,6 +10,11 @@ The automated assurance suite checks that:
 - `mendmark --version`, `--help`, packaged tasks, packaged schemas, and the
   complete single- and multi-agent JSON audit journeys work from that clean
   installation.
+- Harness converters are tested with privacy-safe facsimiles and current public
+  objects from LangChain Core, CrewAI, and OpenAI Agents SDK. `mendmark equip`
+  is tested for idempotency, bounded detection, conflict refusal, symlink
+  containment, explicit expectation approval, and a complete generated
+  evaluator journey.
 - Reports, console output, JUnit, and SARIF do not expose canary values placed in
   prompts, answers, metadata, tags, tool arguments, tool outputs, or descriptions.
 - Repeated audits preserve mutation IDs, ordering, decisions, JUnit, SARIF, and

docs/compatibility.md

@@ -18,9 +18,14 @@ Mendmark 0.x is evolving, but automation still needs predictable contracts.
 - Historical benchmarks select immutable `agent-eval-v1` or `multi-agent-v1`
   mutation profiles. Ordinary audits use `current`; adding a new operator never
   rewrites an older golden-set contract.
+- Harness adapters are dependency-light and use documented public object
+  fields. Hosted CI exercises LangChain Core 1.5.3, CrewAI 1.15.14, and OpenAI
+  Agents SDK 0.19.4 as the 2026-08-10 compatibility snapshot. Newer supported
+  releases are tested through the latest-version harness matrix; Mendmark does
+  not install or pin a harness in customer environments.
 - A custom operator name is globally unique within an audit and is part of the
   customer's accepted baseline contract.
-- `audit`, `audit-json`, `prepare`, `grade`, and `show` exit with 0 for success,
+- `audit`, `audit-json`, `equip`, `prepare`, `grade`, and `show` exit with 0 for success,
   1 for a failed product gate, and 2 for invalid input or infrastructure failure.
 
 The JSON Schemas under `mendmark/schemas` are the machine-readable contract.