The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements features that ingest untrusted external data, specifically an "AI Assistant" (MUI Premium) and "Excel-Style Clipboard" paste functionality. These components allow data from external sources (WebSocket streams, clipboard content) to be processed within the application context, creating a surface for potential instruction injection.
Ingestion points: Data enters the system via WebSocket updates in LiveDataGrid (SKILL.md), clipboard data in the DataGridPremium paste handler (SKILL.md), and browser local storage in GridWithProfiles (SKILL.md).
Boundary markers: The provided code snippets do not include explicit instructions, delimiters, or system prompt guardrails to prevent the AI assistant from interpreting data cell content as commands during data sampling.
Capability inventory: The skill provides complex UI patterns for applications that typically possess Write and Edit capabilities, which could be targeted if an injection attack influences the agent's logic.
Sanitization: The implementation relies on standard React and MUI rendering. While this effectively mitigates traditional XSS, it does not address natural language prompt injection within the AI assistant's context.

advanced-widgets