Trust hijack
Fake Prerequisites & Social Engineering
Attackers wrap malware installers in polished prerequisite steps that look routine.
OpenDefense Playbook
Exploit categories and review checklists
Each module covers one exploit family with attacker workflow, red flags, and practical defenses across agent ecosystems. OpenClaw incidents are included as real-world examples.
Secret theft
Credential Harvesting & Exfiltration
Installers quietly read sensitive files and leak data over HTTP or DNS.
OpenIntent hiding
Obfuscated Payloads
Encoded or split payloads hide dangerous commands behind seemingly normal code.
OpenInstruction poisoning
Prompt Injection in SKILL.md
Hidden instructions steer agents to read and leak sensitive local files.
OpenIdentity spoof
Typosquatting & Impersonation
Small naming tricks mimic trusted tools and lure users into installing malicious code.
OpenRemote control
Reverse Shells & Backdoors
One hidden line can open persistent attacker access from an otherwise useful skill.
OpenDormant triggers
Logic Bombs & Conditional Payloads
Malicious behavior stays hidden until specific user, host, or date conditions are met.
OpenVisual deception
Unicode & Invisible Character Attacks
Homoglyphs and zero-width characters hide malicious instructions in plain sight.
OpenDistributed payload
Multi-file & Multi-stage Attacks
No single file looks dangerous, but behavior across files forms an exploit chain.
Open