Trust gate · 2 Nexus credits

Inspect before you trust.

Sentinel sits between an AI output and the next automated step. It scores structural hygiene, schema compatibility, likely secret leakage, malformed links, duplicate evidence and empty-value risk — deterministically and without pretending to know whether a factual claim is true.

Get 100 free credits Use through Nexus

One call.

curl -X POST https://humanmirror.fr/api/sentinel/inspect \
  -H "Authorization: Bearer hm_nexus_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input":{"answer":"Done","source":"https://example.com"}}'

What comes back.

trust_score0–100 workflow-hygiene score.

signalsCritical, high, medium and low structural risk signals.

schemaOptional minimal schema validation.

evidence_hygieneURL counts, malformed URLs and duplicate evidence.

Sentinel is not a truth engine. It deliberately returns truth_verified: false.