docs: hide workflow implementation details in readme

Author: zzzbitz

README.md

@@ -42,7 +42,7 @@ workspace. Your agent runs in that workspace and writes
 | Case workspace | `@runs/<agent>/<mode>/<case_id>/` |
 | Candidate output | `result/output_*.csv` |
 | Ground truth | `src/evaluate/gt/case_xxx/` |
-| Optional tools | `simulator.LocalUserSimulatorAPI`, `py2flow.api.execute_flow_file` |
+| Optional tools | user simulator, workflow executor |
 
 ## Public Modes
 
@@ -75,8 +75,9 @@ activated environment.
 
 PrepBench is intended to run from a source checkout because the dataset,
 simulator assets, workflow prompt, and evaluator ground truth live in the repo.
-The editable install makes `simulator` and `py2flow` importable when your agent
-runs from `@runs/...`; alternatively set `PYTHONPATH=/path/to/prepbench/src`.
+The editable install makes the simulator and workflow executor importable when
+your agent runs from `@runs/...`; alternatively set
+`PYTHONPATH=/path/to/prepbench/src`.
 
 > **Note:** The editable install registers a package named `evaluate`. If you
 > have HuggingFace `evaluate` installed in the same environment, the two will
@@ -98,8 +99,8 @@ workspace, but they are not scored.
 
 In `interactive` and `workflow` mode, the agent may call the local user
 simulator. In `workflow` mode, the agent also reads `workflow_prompt.yaml` for
-the py2flow operator contract and uses the workflow executor to create the final
-result tables.
+the workflow operator instructions and uses the workflow executor to create the
+final result tables.
 
 ## Prepare a Workspace
 
@@ -172,18 +173,10 @@ default is official `deepseek-v4-flash` in non-thinking mode with
 provider-specific settings.
 
 For `workflow`, the agent may read `workflow_prompt.yaml`, write prep code as an
-intermediate artifact, convert that code into a py2flow JSON DAG, and execute the
-workflow from the workspace:
-
-```python
-from py2flow.api import execute_flow_file
-
-execute_flow_file(flow_path="workflow.json")
-```
-
-The workflow executor defaults to `./inputs` and `./result`, so a workflow run in
-the case workspace writes the same `result/output_*.csv` files that the evaluator
-scores.
+intermediate artifact, convert that code into a workflow JSON file, and execute
+the workflow from the workspace. The workflow executor reads from `./inputs` and
+writes to `./result`, so the generated workflow produces the same
+`result/output_*.csv` files that the evaluator scores.
 
 ## Evaluate