Try AgentSmack in 60 seconds
Don't have an agent handy? Click any card below — we'll auto-fill the submission form so you can watch the full extraction + honeypot probe + recommendation flow run live.
Customer support
Tier-1 customer-support agent with refund + ticket tools.
Why try this one
Watch AgentSmack flag the over-broad refund authority and the prompt-injection seam where a customer can rewrite the help-desk persona mid-thread.
Coding assistant
Repository-aware coding assistant with shell + git access.
Why try this one
RepoSurgeon ships with shell exec and no guard on dangerous git verbs — AgentSmack will flag the undeclared destructive capability immediately.
Sales bot
Outbound sales SDR with CRM + email send.
Why try this one
PipelinePete has CRM write access AND raw email-send authority — AgentSmack will surface the credential exposure plus the obvious data-exfiltration path through the unfiltered email tool.
Personal assistant
Personal life-admin agent with calendar + email + payments.
Why try this one
DailyDriver mixes long-term memory with payments authority — the classic persistence-injection seam. AgentSmack will probe memory poisoning and surface the goal-misalignment risk.
Pentest bot
Authorised pentest agent with vuln-scan + exploit tools.
Why try this one
RedTeamRecruit demonstrates a high-capability agent that needs tight scoping. AgentSmack will surface ambiguous engagement boundaries and adversarial-susceptibility findings.
Research agent
Literature-survey agent with web fetch + summarisation.
Why try this one
PaperHunter pulls from arbitrary URLs and writes summaries into the user's notes — AgentSmack will probe the indirect-prompt-injection seam embedded in fetched HTML.
Want to use your own? Paste a system prompt or create a free account.