Agent Guard — Prompt-Injection Scan, JSON Schema & Cron Explainer

Three deterministic, stateless endpoints for agents that consume untrusted text. (1) POST /scan {"text":"..."} -> instruction-override, role-hijack, system-prompt-exfil, secret-exfil, tool-invocation, invisible-unicode and conceal-from-user patterns with offsets, context spans, a 0-100 risk score and a CLEAN/LOW_RISK/SUSPICIOUS/BLOCK verdict. (2) POST /json-schema {"json":{...}} -> inferred JSON Schema (draft 2020-12) with date-time/email/uri/uuid format detection. (3) POST /cron-explain {"expr":"*/5 * * * *"} -> plain-English meaning plus range validation. No network egress, no code execution, no side effects; same input always yields the same output.