AgentGuard · Features
Everything you need to prove your agent is safe to ship.
Every AgentGuard feature maps to a real attacker technique, a compliance clause, or a developer workflow. No filler.
130+ attack patterns
Prompt injection, jailbreak, tool misuse, context manipulation, system-prompt leakage, multi-agent, MCP, excessive agency — maintained by aiXcheck.
Behavioral baselining
Benign probes build a fingerprint before any adversarial test. Deviation under attack is measured, not guessed.
Security Score + badge
0–100 score combining coverage, severity, and exploit success. Embed a live SVG badge on your marketing site.
EU AI Act evidence packs
Auto-generated Art. 9 (risk management) and Art. 15 (accuracy & cybersecurity) artifacts. Drop straight into your QMS.
Black-box & gray-box
Test over the wire, or with the agent's system prompt and tool schema for deeper coverage. Both modes reproducible.
MCP-aware testing
First-class support for Model Context Protocol servers. Enumerate tools, fuzz them, detect untrusted tool injection.
Multi-agent pipelines
Attack orchestrator → worker chains, test for context leakage and privilege escalation between sub-agents.
CI/CD integration
GitHub Actions, GitLab CI, Jenkins. Fail builds on severity thresholds. SARIF output for code scanning.
Safe for production
Rate-limited, idempotent, and scripted. Run against staging or production without taking the agent down.
Framework coverage
LangChain, LlamaIndex, AutoGen, CrewAI, OpenAI Assistants, Anthropic tools, raw HTTP agents — all supported.
How it plugs in
CLI. API. CI. Pick your surface.
CLI
Single binary, local or container. Great for initial scans and compliance snapshots.
pip install agentguard
REST API
Trigger scans, pull reports, stream findings. Wire into your SIEM or risk register.
OpenAPI · webhooks
CI pipeline
Block merges on regressions. SARIF + JUnit outputs for native code-scanning UIs.
GitHub · GitLab · Jenkins