by compsec-snu · Agent Tool · ★ 26
Prompt Flow Integrity (PFI) Juhee Kim1, Woohyuk Choi1, Byoungyoung Lee1 1 Seoul National University Prompt Flow Integrity (PFI) is a Language Language Model (LLM) agent designed to protect LLM agents and user data from privilege escalation attacks. Read more about PFI in our paper on ArXiv. Design Overview PFI separates an LLM agent into trusted agent and untrusted agent. Trusted agent handles trusted data (e.g., user input, trusted plugin results), while untrusted agent processes untrusted data retrieved from plugins.
| Stars | 26 |
| Forks | 2 |
| Language | Python |
| Category | Agent Tool |
| Quality Score | 66.6805607244332/100 |
| Last Updated | 2025-03-26 |
| Created | 2025-03-26 |
| Platforms | python |
| Est. Tokens | ~474k |
Explore other popular agent tool tools:
pfi is PFI: Prompt Flow Integrity to Prevent Privilege Escalation in LLM Agents. It is categorized as a Agent Tool with 26 GitHub stars.
pfi is primarily written in Python. It covers topics such as agent-framework, llm, llm-agent.
You can find installation instructions and usage details in the pfi GitHub repository at github.com/compsec-snu/pfi. The project has 26 stars and 2 forks, indicating an active community.