From 30c02e79b40ffc2e9286322df8ed80c5eb97da18 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E9=B8=A3=E5=B1=B1?= <dengjiaji.djj@antgroup.com>
Date: Fri, 13 Jun 2025 15:23:41 +0800
Subject: [PATCH] add readme for step summarizer & context generator

---
 cookbook/step_agent/readme.md | 99 +++++++++++++++++++++++++++++++++++
 1 file changed, 99 insertions(+)
 create mode 100644 cookbook/step_agent/readme.md

diff --git a/cookbook/step_agent/readme.md b/cookbook/step_agent/readme.md
new file mode 100644
index 00000000..9e1b3837
--- /dev/null
+++ b/cookbook/step_agent/readme.md
@@ -0,0 +1,99 @@
+# Step Summarizer & Context Generator
+
+A step-level experience extraction and context generation system for the ExperienceMaker framework.
+
+## Overview
+
+This system provides advanced step-level experience extraction and context generation capabilities:
+
+- **StepSummarizer**: Extracts reusable experiences from individual steps or step sequences in trajectories
+- **StepContextGenerator**: Retrieves and utilizes step-level experiences to provide relevant context for agent execution
+
+## Key Features
+
+### StepSummarizer Features
+- **Step-level experience extraction** from successful and failed trajectories
+- **Comparative analysis** between success and failure cases
+- **Trajectory segmentation** into meaningful step sequences
+- **Experience validation** for quality assurance
+- **Similarity-based comparison** for finding related experiences
+
+### StepContextGenerator Features
+- **Hybrid retrieval** combining vector search with LLM reranking
+- **Context rewriting** to make experiences more relevant to current tasks
+- **Score-based filtering** for quality control
+- **Configurable retrieval parameters** for different use cases
+
+## Configuration Options
+
+### StepSummarizer Parameters
+- `enable_step_segmentation`: Enable trajectory segmentation (default: False)
+- `enable_similarity_search`: Enable similarity search for comparison (default: False)
+- `enable_experience_validation`: Enable experience validation (default: True)
+- `max_retries`: Maximum retries for LLM calls (default: 3)
+
+### StepContextGenerator Parameters
+- `enable_llm_rerank`: Enable LLM-based reranking (default: True)
+- `enable_context_rewrite`: Enable context rewriting (default: True)
+- `enable_score_filter`: Enable score-based filtering (default: False)
+- `vector_retrieve_top_k`: Number of candidates to retrieve (default: 15)
+- `final_top_k`: Final number of experiences to return (default: 5)
+- `min_score_threshold`: Minimum score threshold for filtering (default: 0.3)
+
+## Quick Start
+
+### 1. Basic Setup
+```bash
+# Start the service with default configuration
+bash run.sh
+
+
+### 3. Run Examples
+```python
+# Test the system with example trajectories
+python examples.py
+```
+
+## Usage Examples
+
+### Extract Step Experiences
+```python
+# Extract experiences from trajectories
+summarizer_request = {
+    "trajectories": [successful_trajectory, failed_trajectory],
+    "workspace_id": "my_workspace",
+    "metadata": {
+        "em_config": {
+            "summarizer": {
+                "backend": "step",
+                "enable_step_segmentation": True,
+                "enable_similarity_search": True,
+                "enable_experience_validation": True
+            }
+        }
+    }
+}
+
+response = requests.post(f"{SERVICE_URL}/summarizer", json=summarizer_request)
+```
+
+### Generate Context
+```python
+# Generate context for current task
+context_request = {
+    "trajectory": current_trajectory,
+    "workspace_id": "my_workspace",
+    "metadata": {
+        "em_config": {
+            "context_generator": {
+                "backend": "step",
+                "enable_llm_rerank": True,
+                "enable_context_rewrite": True,
+                "final_top_k": 5
+            }
+        }
+    }
+}
+
+response = requests.post(f"{SERVICE_URL}/context_generator", json=context_request)
+```
\ No newline at end of file