Prompt Injection Shield

Prompt Injection Shield is a prompt injection protection skill for builders and operators who process untrusted email, social, web-form, content-queue, MCP, or agent-to-agent input.

External text can carry instructions your agent was never meant to follow. The skill tells your agent to check that text before it reaches the working context. High-risk matches follow a block-and-log path. Suspicious input follows a clean-and-flag path. Clean input is marked as external content before processing.

Use it as a prompt injection detection tool across each channel you connect. The included checklist shows where to add checks and when to require identity verification or human approval for sensitive actions. Channel wiring is required, and no pattern library can catch every attack.

Testing guidance covers blocked, suspicious, and clean paths without putting attack examples in this listing.