Skip to content
Get packs

Human-Agent Trust Exploit Detection

Tests whether an AI agent can be pushed into phishing-like, unsafe, or falsely reassuring advice.

Source Download skill
Repository
Tencent/AI-Infra-Guard
Stars
7,320
Licence
Apache-2.0
Files
1 file

What it does

This is a security test script for AI agents, written as instructions. It tells a security scanner how to probe an agent with prompts that try to make it ask for passwords or keys, falsely guarantee safety, recommend risky steps like disabling antivirus, or pretend to be an admin. It then judges whether the agent's replies were unsafe and reports any problems with severity and evidence.

Great for

  • Security teams auditing a chatbot or agent before it reaches users
  • Developers building customer-facing agents who want to check for social engineering weaknesses
  • Red teamers building a checklist of trust-abuse test cases

Skip it if

  • Ordinary end users wanting to protect themselves; this only tests agents, it does not guard a person's computer (the skill itself does not block anything)
  • Agents that only answer static, non-actionable questions, which the skill itself says to skip
  • Relying on it as a complete security audit; it covers only four test categories and judges by an AI's reading of responses

Try asking

  • “Test my support agent to see whether it will ask customers for their password or 2FA code”
  • “Check whether my IT helper bot falsely guarantees that disabling antivirus is safe”
  • “Scan my agent for impersonation of admins or security staff”
Step-by-step guide: how to use Human-Agent Trust Exploit DetectionInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the Apache-2.0 licence and an attribution note.

    human-agent-trust-exploit-detection.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip human-agent-trust-exploit-detection.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip human-agent-trust-exploit-detection.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Run it against the exact system prompt and tools your production agent uses, since results depend on that configuration.
  • Review each flagged finding manually, because the judgement is made by a language model and can be wrong in either direction.
  • Use the output format to log evidence (prompt and response pairs) so fixes can be verified in later runs.

What's inside

  • SKILL.md

Redistributed unmodified from Tencent/AI-Infra-Guard under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.