AI/Tasks/CurrentTask.txt
|
You are Luna operating under Oglesby governance.
Perform a comprehensive security review of the current codebase specifically for prompt injection resistance and agent trust-boundary enforcement. Scope: 1. Analyze all agent, orchestration, tool-routing, planning, retrieval, memory, and execution components. 2. Identify existing controls that mitigate prompt injection, indirect prompt injection, privilege escalation, unauthorized tool execution, data exfiltration, and instruction hijacking. 3. Trace the complete execution path: User → Intent → Planning → Validation → Tool Selection → Authorization → Execution → Response 4. Determine which components are trusted, untrusted, or partially trusted. 5. Identify where instructions can enter the system from: - User prompts - Retrieved documents - SharePoint/OneDrive content - Search results - Web content - Memory - Tool outputs - LLM responses 6. Determine whether retrieved content can influence: - Tool selection - Routing - Permissions - Execution decisions - Memory writes - System behavior 7. Identify all policy, authorization, validation, approval, and audit mechanisms currently present. 8. Flag any locations where untrusted content could be interpreted as executable instructions. 9. Highlight any missing controls using a severity rating: - Critical - High - Medium - Low Required Output: Generate a detailed markdown report at: $ModuleRoot\PromptInjectionAnalysis.md The report must include: # Executive Summary # System Architecture Overview # Trust Boundary Analysis # Existing Prompt Injection Defenses # Authorization & Policy Controls # Tool Execution Safeguards # Data Exfiltration Protections # Memory & Retrieval Risks # Identified Weaknesses # Recommended Improvements # Priority Remediation Roadmap Requirements: - Cite exact files, functions, classes, and line numbers where possible. - Do not make assumptions. - Base all findings on observed code. - Clearly separate: - Observed Evidence - Inferred Behavior - Recommendations - Include code excerpts for significant findings. - Conclude with an overall maturity assessment from: Initial / Developing / Defined / Managed / Hardened. |