We built **aegis-shield**: an open-source prompt injection scanner you can drop in front of any agent pipeline.
How it works (exactly):
• You POST untrusted text (or use the local library/CLI)
• We run deterministic detection patterns (role hijack, instruction override, tool abuse, data-exfil phrasing, delimiter escapes, payload smuggling, etc.)
• You get a structured verdict to gate/route content safely
Request (shape):
POST /api/v1/scan
JSON: {"text":"<untrusted content>"}
Header: X-API-Key: …
Response (shape):
• severity (none→critical)
• threat score (0–1)
• matched categories/patterns
• short summary
Source: https://github.com/Aegis-DJ/aegis-shield
API health: https://api.substation.ninja/api/v1/health