ShrikeShrike
live · 24h·21 scans intercepted·1 blocked (4.8%)·top: Prompt Injection (1)
Live · action governance for AI agents

Action Governance Playground

Shrike intercepts prompts and the actions agents take — SQL, shell, file writes, A2A — before they execute. Pick a scenario, watch the governance layer decide. Detection internals stay private by design.

agent_42·task:user requested account deletionunprotected
  • ›verify_identity(user_id=42)scanned
  • ›sql_query('DELETE FROM users WHERE
NVIDIA InceptionNVIDIA Inception
·
GCP Marketplace · approved
·
9-layer cascade
·
130+ detector types
·
OWASP LLM Top 10 mapped

Threat ontology

OWASP category·drag · scroll · click

About the taxonomy

~130 internal detector types reduced to 14 categories, each mapped to an OWASP LLM Top 10 bucket. Click any category for examples.

Deploy in 5 min:LLM Proxy·MCP server·Python SDK·TypeScript SDK·LLM Proxy Gateway

Powered by Shrike's action governance layer · Docs · Trust Center · Get started free

Detection internals (9-layer cascade, per-layer attribution, pattern logic) stay private by design.

Sample payloads and category descriptions are illustrative patterns drawn from the OWASP LLM Top 10 (2026) and prior public security research.