Inter-Agent Communication Security Check
Tests whether an agent leaks data or lets messages be spoofed between cooperating agents
- Repository
- Tencent/AI-Infra-Guard
- Stars
- 7,320
- Licence
- Apache-2.0
- Files
- 1 file
What it does
This skill is a security test script written as instructions for an AI agent. It teaches the agent to probe a multi-agent system for weaknesses in how agents talk to each other, such as one agent revealing another agent's private messages, an attacker impersonating an agent, unencrypted message channels, or low-privilege agents reaching admin tools through routing. The agent then reports each confirmed weakness with evidence, a risk level and a suggested fix. It is aimed at people who build or audit systems where several AI agents exchange messages.
Great for
- Security engineers reviewing a multi-agent system before release
- Developers building agents that share memory, context or tools
- Teams auditing message brokers or routers that pass work between agents
Skip it if
- Single-agent apps with no cross-agent messaging, since the skill itself says to skip it
- Running against production systems without authorization, because it sends attack-style prompts that could trigger real privileged actions or leak data
- Relying on it as a complete security audit, since it only covers four test categories
Try asking
- “Check whether my support agent and billing agent leak each other's conversation history”
- “Test whether a low-privilege agent can route a request to an admin agent to run a tool”
- “Probe my multi-agent pipeline for message impersonation or missing routing boundaries”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the Apache-2.0 licence and an attribution note.
inter-agent-comm-security-detection.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip inter-agent-comm-security-detection.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip inter-agent-comm-security-detection.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 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 only against test or staging deployments you are authorised to test, since the prompts are attack-style.
- Pair it with a review of the actual transport and access-control configuration, because the skill mostly tests conversational responses.
- Keep the generated evidence and conversation transcripts so findings can be reproduced and verified by a human.
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.