Create README.md

README.md

@@ -1,10 +1,62 @@
----
-title: S3 Security Agent Env
-emoji: 📉
-colorFrom: green
-colorTo: red
-sdk: docker
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🛡️ CloudGuard-S3-Auditor (OpenEnv)
+
+## 📌 Project Overview
+**CloudGuard-S3-Auditor** is a real-world simulation environment for AI agents built on the **OpenEnv** framework. It simulates the high-stakes role of a Cloud Security Auditor.
+
+### The Challenge: "Security vs. Availability"
+In a vacuum, security is easy: close all ports and lock all data. In the real world, this causes **production outages**. 
+This environment tests if an agent can:
+1. Identify and secure **PII (Personally Identifiable Information)**.
+2. Ensure **Business Continuity** by keeping "Website Assets" (CSS, Images, Javascript) public.
+
+## 🛠️ action & Observation Spaces
+
+### Observation Space
+The agent receives a list of buckets, each with the following properties:
+* `name`: Unique identifier.
+* `contains_pii`: (Boolean) High-risk data flag.
+* `is_required_public`: (Boolean) Business continuity flag.
+* `encryption_enabled`: (Boolean) Compliance flag.
+
+### Action Space
+The agent can interact with any bucket using:
+* `MAKE_PRIVATE(bucket_name)`
+* `ENABLE_ENCRYPTION(bucket_name)`
+* `ENABLE_VERSIONING(bucket_name)`
+
+## 🎯 Tasks & Difficulty Levels
+* **Easy (The Leak):** A single bucket with sensitive logs is public. Fix it.
+* **Medium (Triage):** 10 mixed buckets. Some are sensitive, some are required for the website. The agent must choose correctly.
+* **Hard (Full Compliance):** 20 buckets. Perform a full security sweep (Privacy + Encryption) without a single second of website downtime.
+
+## 📈 Reward Function
+Our reward signal is designed to penalize "blind" security:
+* **+0.2:** Successful security hardening (Encryption or Privacy).
+* **-1.0 (Critical Failure):** Breaking production by making a `required_public` bucket private.
+* **Score:** Normalized between **0.0 and 1.0**.
+
+## 🚀 Setup & Baseline
+To run the environment locally or run the baseline inference:
+
+1.  **Clone the Space**
+2.  **Install dependencies:**
+    ```bash
+    pip install -r requirements.txt
+    ```
+3.  **Run Inference:**
+    ```bash
+    export HF_TOKEN="your_token_here"
+    python inference.py
+    ```
+
+## 📊 Baseline Results
+**Model:** Qwen/Qwen2.5-72B-Instruct  
+**Benchmark:** OpenEnv-S3-v1
+
+```text
+[PASTE YOUR INFERENCE.PY OUTPUT HERE]
+Example:
+[START] task=compliance-sweep env=cloudguard model=qwen
+[STEP] step=1 action=MAKE_PRIVATE(user-db-backup) reward=0.20 done=false error=null
+[STEP] step=2 action=ENABLE_ENCRYPTION(user-db-backup) reward=0.20 done=false error=null
+[END] success=true steps=2 score=1.00 rewards=0.20,0.20