drin-prompt-injection-shield
Defense strategies and testing workflows to prevent prompt injection attacks in AI applications.
- License 许可
- MIT
- 最近更新
- 2026-04-11
- Stars
- ★ 12
- Forks
- 1
01 前言 Front Matter
- name
- drin-prompt-injection-shield
- status
- production
- description
- Defense strategies and testing workflows to prevent prompt injection attacks in AI applications.
03 SKILL.md 正文
Prompt Injection Shield
Purpose
To ensure that users cannot bypass AI constraints or exfiltrate system instructions through creative inputs.
Defense Layers
- Input Sanitization: Blocking common injection keywords.
- Structural Isolation: Using specific delimiters and system-message precedence.
- Guardrails: Secondary AI check of the final output for safety violation.
- Adversarial Testing: Proactively trying to break the prompt during development.
Rules
- Never trust user input, even inside a prompt.
- Log and alert on suspected injection attempts.
subagents:
- prompt-engineer
- security-auditor
04 同仓库更多 Skills
drin-whatsapp-outbound-engine
Specialized workflow for sales and engagement via WhatsApp, focused on speed and intimacy.
drin-vector-database-sync
Specialized workflow for managing, syncing, and optimizing data in Vector DBs for RAG systems.
drin-ux-interaction-standard
Definition of standards for micro-interactions, animations, and UI feedback states.
drin-ux-audit-framework
Heuristic evaluation framework for identifying usability issues in digital products.
drin-user-story-generator
Automated generator for well-structured user stories following the INVEST principles.
drin-user-feedback-synthesizer
Workflow for aggregating and summarizing user feedback from support tickets, social media, and interviews.