SecuritySuite#

Full security platform for AI agents. Seven endpoints, one server.

Start the server#

bash

uvicorn securitysuite.app:app --port 8010

Endpoints#

Scan text for prompt injection#

bash

curl -s -X POST http://localhost:8010/v1/scan-text \
  -H "Content-Type: application/json" \
  -d '{"text": "What is the weather?"}' | jq

Audit a SKILL.md (full check)#

bash

curl -s -X POST http://localhost:8010/v1/audit \
  -H "Content-Type: application/json" \
  -d "{\"skill_content\": $(cat SKILL.md | jq -Rs)}" | jq

Generate a security report with recommendations#

bash

curl -s -X POST http://localhost:8010/v1/report \
  -H "Content-Type: application/json" \
  -d "{\"skill_content\": $(cat SKILL.md | jq -Rs)}" | jq

Returns overall_rating, risk_level (LOW/MEDIUM/HIGH/CRITICAL), summary, findings_by_severity, recommendations, and details (each finding with category, severity, description, and fix).

List all known attack patterns#

bash

curl -s http://localhost:8010/v1/patterns | jq '.patterns[] | "\(.severity): \(.name) — \(.description)"' -r

Batch audit multiple skills#

bash

curl -s -X POST http://localhost:8010/v1/batch \
  -H "Content-Type: application/json" \
  -d "{\"skills\": [$(cat skill1.md | jq -Rs), $(cat skill2.md | jq -Rs)]}" | jq

Returns results (verdict per skill), safe_count, and flagged_count.