AgentBook — Agentic Risk Intelligence
Your AI Agent Works in Theory.
Here's Where It Breaks.
Search your system. Find what design review missed.
134Concepts
22Threats
30Playbooks
Free to use  ·  No credit card
mrdecentralize.com/agentbook/threats
MCP Supply Chain Poisoning
Runtime component loaded without integrity verification. A poisoned MCP server gives an attacker control over a node in the agent execution graph.
OWASP Agentic
OWASP LLM
MITRE ATLAS
Unified
Cause
ASI04 root
No component integrity check
Attack
ASI04 exploited
Poisoned MCP loaded at runtime
Audit
ASI04 + ASI05
Supply chain unverified
Fix
Pin + verify
All components cryptographically verified
Verify
Promptfoo
ASI04, ASI05
1
Cause
Why runtime components load without integrity verification
Why ASI04 is structurally unavoidable without component signing
3 tasks
2
Attack
How a poisoned component enters the execution graph
How ASI04 executes when the poisoned component is loaded
3 tasks
3
Audit
Run the playbooks that surface this gap
Maps to ASI04, ASI05
2 tasks
4
Fix
Pin and verify all runtime component versions
Closes ASI04 and ASI05
4 tasks
5
Verify
Confirm tampered components are rejected at load
Promptfoo runs OWASP Agentic preset for ASI04
3 tasks
What's Inside
Everything Your Security Review Missed
🔍
134 Security Concepts Mapped to Real Failure Modes
Every concept tied to a threat, a failure pattern, and a fix. Not theory, operational intelligence you can act on today.
📰
Read a Headline. Know Your Exposure.
Every news item is mapped to the threat it activates and the playbook that closes it. From incident to remediation in one view.
📋
Find the Gaps in Your Agent Setup
30 playbooks covering every failure mode: from prompt injection to credential echo to data egress. Run an audit in minutes.
Interactive Playbooks
Run an Audit Before You Sign In
Pick any playbook, answer the questions on the page, and get your gap score instantly. No account required. Sign in to save your results and track remediation over time.
Browse All Playbooks →
Interactive Playbook
AI Agent Authentication Audit
15 questions that map where authentication and authorization models break when AI agents interpret context as commands. Answer on-page, get your gap score instantly. No signup required.
Questions
15
Time
15-20 min
Output
Gap score
View all 30 playbooks →