refactor(harness)!: post-merge cleanup + init templates for the unified harness surface (PR 10)

.github/workflows/harness-integration.yml

@@ -7,8 +7,7 @@ on:
     paths:
       - "src/agentex/lib/core/harness/**"
       - "src/agentex/lib/adk/_modules/**"
-      - "tests/lib/core/harness/test_harness_pydantic_ai_*.py"
-      - "tests/lib/core/harness/test_harness_langgraph_*.py"
+      - "tests/lib/core/harness/test_harness_*.py"
       - ".github/workflows/harness-integration.yml"
 
 jobs:
@@ -34,14 +33,15 @@ jobs:
         run: ./scripts/test tests/lib/core/harness/ -v
 
   # Offline harness integration tests (sync / async / temporal channels) for each
-  # migrated harness. These use fake streams / TestModel + fake streaming/tracing
-  # and require no live infrastructure. Future harness migration PRs (6-8) add
-  # their harness to the matrix below and their test paths to the triggers above.
+  # harness. These use fake streams / TestModel + fake streaming/tracing and
+  # require no live infrastructure. All five harnesses are now covered; the
+  # trigger above uses a `test_harness_*.py` glob so new suites are picked up
+  # automatically.
   live-matrix:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        harness: [pydantic_ai, langgraph]
+        harness: [pydantic_ai, langgraph, openai, claude_code, codex]
         channel: [sync, async, temporal]
       fail-fast: false
     name: ${{ matrix.harness }}-${{ matrix.channel }}

CHANGELOG.md

@@ -2,6 +2,11 @@
 
 ## Unreleased
 
+### ⚠ BREAKING CHANGES
+
+* **harness:** removed the deprecated bespoke tracing handlers `create_langgraph_tracing_handler` / `create_pydantic_ai_tracing_handler` (and their `AgentexLangGraphTracingHandler` / `AgentexPydanticAITracingHandler` classes) from the public `agentex.lib.adk` surface. Span tracing is now derived from the canonical `StreamTaskMessage*` stream by `UnifiedEmitter` — wrap your run in the harness `*Turn` and drive `UnifiedEmitter.yield_turn` / `auto_send_turn`. The `agentex init` templates were migrated accordingly.
+* **harness:** each harness now exposes exactly `_<harness>_sync.py` + `_<harness>_turn.py` under `agentex.lib.adk._modules`. The OpenAI harness `OpenAITurn` and `convert_openai_to_agentex_events` moved to `agentex.lib.adk._modules._openai_turn` / `_openai_sync`; back-compat shims remain at `agentex.lib.adk.providers._modules.{openai_turn,sync_provider}` for one release. Public facade names (`stream_pydantic_ai_events`, `stream_langgraph_events`, `emit_langgraph_messages`, etc.) are unchanged.
+
 ### Features
 
 * **tracing:** emit OTel metrics for async span queue depth, batch drain, and SGP export success/failure (HTTP status labels). Disable SDK-side recording with ``AGENTEX_TRACING_METRICS=0``.

adk/docs/harness.md

@@ -39,14 +39,17 @@ Every harness tap produces a sequence of these. Everything downstream (delivery,
 
 ## Per-harness taps: `convert_<harness>_to_agentex_events`
 
-A tap is an async generator that translates the harness's native event stream into `StreamTaskMessage*` events. The currently shipped taps are:
+A tap is an async generator that translates the harness's native event stream into `StreamTaskMessage*` events. The shipped taps are:
 
 | Harness | Tap function | Exported from |
 |---|---|---|
 | pydantic-ai | `convert_pydantic_ai_to_agentex_events` | `agentex.lib.adk` |
 | LangGraph | `convert_langgraph_to_agentex_events` | `agentex.lib.adk` |
+| claude-code | `convert_claude_code_to_agentex_events` | `agentex.lib.adk` |
+| codex | `convert_codex_to_agentex_events` | `agentex.lib.adk` |
+| OpenAI Agents | `convert_openai_to_agentex_events` | `agentex.lib.adk.providers._modules.sync_provider` |
 
-Taps for claude-code and codex will be added in subsequent PRs (AGX1-420, AGX1-421) and exported from `agentex.lib.adk` in the same way.
+Each harness also provides a `HarnessTurn` wrapper that pairs its tap's event stream with usage extraction: `PydanticAITurn`, `LangGraphTurn`, `ClaudeCodeTurn`, `CodexTurn`, and `OpenAITurn`.
 
 ---
 
@@ -157,11 +160,13 @@ Spans are derived from the canonical stream by `SpanDeriver` (pure, no `adk` dep
 
 ## Usage examples by channel
 
-### Sync ACP (pydantic-ai tap)
+### Sync ACP (`yield_turn`)
+
+Build the harness's `HarnessTurn` wrapper and iterate `emitter.yield_turn(turn)` — the emitter forwards each event to the caller and traces spans as a side effect:
 
 ```python
 import agentex.lib.adk as adk
-from agentex.lib.adk import UnifiedEmitter, convert_pydantic_ai_to_agentex_events
+from agentex.lib.adk import UnifiedEmitter, ClaudeCodeTurn
 
 @acp.on_message_send
 async def handle(params):
@@ -172,13 +177,12 @@ async def handle(params):
             trace_id=task_id,
             parent_span_id=turn_span.id if turn_span else None,
         )
-        tap = convert_pydantic_ai_to_agentex_events(pydantic_stream)
-        # wrap tap in a HarnessTurn then yield_turn, or yield directly:
-        async for event in tap:
+        turn = ClaudeCodeTurn(claude_code_stream)   # any HarnessTurn
+        async for event in emitter.yield_turn(turn):
             yield event
 ```
 
-For the pre-unified sync path the tap is still yielded directly; `UnifiedEmitter.yield_turn` is the forward-looking integration point when a `HarnessTurn` wrapper is available.
+Every harness follows the same shape — swap `ClaudeCodeTurn` for `PydanticAITurn`, `LangGraphTurn`, `CodexTurn`, or `OpenAITurn` and feed it that harness's native stream.
 
 ### Async Temporal (auto-send)