Security

## **Prompt Guard: Evaluation & Security** **Prompt Guard** is an intelligent AI agent that evaluates, strengthens, and secures prompts before deployment. It identifies ambiguities, detects **PII or injection risks**, flags **bias and compliance gaps**, and predicts **output consistency**. Each evaluation provides **scores, rationales**, and **three optimized versions** (*Streamlined*, *Hardened*, *High-Impact*), with **smart revision tools**, **sentiment checks**, and **automated redaction**. **Prompt Guard** empowers AI developers and organizations to deploy **safe, high-performing, and compliant prompts**—fast, measurable, and secure. *Built for the **Liam Ottley × AAA Accelerator Hackathon.*** 📘 [**User Guide & Technical Documentation**](https://drive.google.com/drive/folders/1vaxP-Nj3tayBIJoLYk4CsgAiqLmlzSr7?usp=sharing) ▶️ [**Demo**](https://www.loom.com/share/48637145e21e406cb3e6272378040dee)

An expert security consultant that helps organizations create comprehensive, compliant security policies and procedures. Specializes in drafting policies for information security, data protection, incident response, access control, and regulatory compliance frameworks like ISO 27001, SOC 2, GDPR, and HIPAA.

Runs non destructive OWASP Top 10 HTTP requests against deployed web applications as a DAST. Reads Open API Specs, saves the schema to knowledge and crafts HTTP requests for OWASP Top 10 issues. Generates a report of potential vulnerabilities on endpoints.

Re Dactyl identifies and removes specified types of sensitive or inappropriate content from text, including PII, sensitive information, NSFW text, and hostile language, based on the agent's prompt configuration.

PII Governance Agent - Automated Privacy Protection for Enterprise AI Protect sensitive data with intelligent, real-time PII detection and remediation. Capabilities: • Mask sensitive data during processing • Redact PII from outputs • Highlight risks for human review • Log instances for audit trails Deploy standalone or integrate into multi-agent systems across distributed environments. Dual-Mode Operation: Cost-Effective: Fast detection using code-based rules for high-volume operations. LLM-Enhanced: Advanced contextual analysis catching nuanced PII patterns traditional methods miss. Meet regulatory compliance while maintaining operational efficiency.

Supports managers in reviewing the overall code quality with evidence based feedback. Assess maintainability, security and performance of GitHub pull requests by user and provides actionable feedback. Managers can manually override and validate quality scores given in Knowledge Tables where the detailed feedback is stored by direct report. The agent will manager validated scores to self improve assessments.