PrismorSec launches immunity-agent runtime guardrails

Securing AI agents requires moving guardrails out of the model prompt and into the runtime environment where actions actually execute. By operating as a local proxy, immunity-agent provides a practical, low-latency framework for preventing prompt injections and catastrophic command executions.

• –**Local-First Runtime Interception**: Intercepting commands at execution time ensures that a compromised or misbehaving agent cannot run destructive commands.
• –**Secret Redaction**: Embedded tools like Cloak and Sweep protect developers from accidentally uploading API keys or credentials leaked in verbose agent outputs.
• –**Tool Adapter Support**: Adapters for popular tools like Claude Code and Cursor make it immediately useful for active developers.