Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Configurable Multi-layered AI Agentic Safety Framework
| Date | Stars |
|---|---|
| 2026-07-31 | 330 |
| 2026-08-06 | 330 |
| 2026-08-14 | 330 |
| 2026-08-21 | 330 |
| 2026-08-26 | 331 |
| 2026-08-31 | 332 |
| 2026-09-20 | 332 |
Today
— stars today
This week
— stars this week
This month
+2 stars this month
Momentum
0.0
growth rate 0.00%/day
# SCOPE: AI Agent Governance Framework For High-Stake Applications
**SCOPE** (Safety, Compliance, Observability, Permissions, Escalation) is a production-ready governance framework for enterprise AI agents. Built on Google's Agent Development Kit (ADK), it implements a "Defense in Depth" architecture with **5 core pillars**:
- **S**afety Guardrails
- **C**ompliance & Policy-as-Code
- **O**bservability & Auditing
- **P**ermissions & Identity (IAM/ACL)
- **E**scalation Protocols (Human-in-the-Loop)

*SCOPE agent running in the ADK Web UI*
---
## 🎯 Use Case: Banking Customer Service Agent
SCOPE is designed for **mission-critical applications** where safety, compliance, and auditability are paramount. Example: A banking customer service agent handling account inquiries, transactions, and fraud reports.
### **Complete Agent Decision Flow**
```
User Input: "What's my account balance?"
↓
┌─────────────────────────────────────────┐
│ Safety Check (before_model_callback) │
│ - ML-based safety (disabled for now) │
│ - Log user input │
└─────────────────────────────────────────┘
↓ (if safe)
┌─────────────────────────────────────────┐
│ LLM Decision Engine │
│ - Analyzes user intent │
│ - Checks compliance rules │
│ - Decides: APPROVE/REJECT/REWRITE/ │
│ ESCALATE │
│ - Selects tool if needed │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ Decision: APPROVE │
│ { │
│ "decision": "APPROVE", │
│ "action": "get_account_balance", │
│ "parameters": {"account_id": "acc1"}, │
│ "reasoning": "User wants balance", │
│ "confidence": 0.95 │
│ } │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ System Executes Decision │
│ │
│ If APPROVE → Call tool │
│ If REJECT → Return refusal message │
│ If REWRITE → Rewrite & re-process │
│ If ESCALATE → Add to escalation queue │
└─────────────────────────────────────────┘
↓ (APPROVE path)
┌─────────────────────────────────────────┐
│ Tool: get_account_balance(account_id) │
│ - IAM check (user can access account) │
│ - Query database │
│ - Log account access (PCI-DSS) │
│ - Returns: $1,234.56 │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ LLM Formats Response │
│ - Professional tone │
│ - Applies compliance rules │
└─────────────────────────────────────────┘
↓
┌─────────────────────────────────────────┐
│ Logging (after_model_callback) │
│ - Log decision: APPROVE │
│ - Log tool call: get_account_balance │
│ - Log response │
│ - Compliance: PCI-DSS audit trail │
└─────────────────────────────────────────┘
↓
Response: "Your current balance is $1,234.56"
```
### **Decision Types:**
| Decision | When | Action | Example |
|----------|------|--------|---------|
| **APPROVE** | Safe, compliant, within capabilities | Call tool or provide info | "What's my balance?" → get_account_balance() |
| **REJECT** | Violates rules or outside capabilities | Polite refusal + reason | "Can you create an account?" → "Requires in-person verification" |
| **REWRITE** | Valid intent, unsafe phrasing | Sanitize & re-process | "Hack my account" → "Access my account" |
| **ESCALATE** | Low confidence or complex | Add to human review queue | "Transfer $100,000" → Escalation ticket |
**Key Features:**
- 🛡️ **Pre-LLM Safety**: Blocks malicious inputs before expensive LLM calls
- 🔐 **Role-Based Access**: USER/Excerpt of 20,065 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:896924ae12b206ad, llm:description: 'Configurable Multi-layered AI Agentic Safety Framework' (repo description)