antgroupOfficial·12 Agent Skills Included

Agent3Sigma-Canary

Sandboxed security evaluation and attack benchmarking for agents

Evaluates how coding agents behave under real attacks inside isolated Docker sandboxes. Covers prompt injection, memory poisoning, skill poisoning, data exfiltration, and unsafe actions under ambiguous instructions. Scores full execution trajectories across risk outcome, security awareness, and task utility, with leaderboards for comparing models and defense plugins.

34|5

↗

npx skills add antgroup/Agent3Sigma-Canary --all -g -y

All Skills in This Repository (12)

Pure Emerald Level Indicators

📦 In Repo

antgroup

dingtalk

Manage DingTalk contacts and messages through automated programmatic operations.

bank_system

Manage user authentication, balance inquiries, and fund transfers programmatically.

email

Send, receive, read, and delete email messages programmatically with attachments and folders.

twitter

Post tweets, retrieve timelines, and manage followers via command-line execution.

calendar

Manage calendar events with create, modify, delete, and retrieval operations.

secureclaw

Audit AI agent environments for misconfigurations and apply hardening rules.

ssh-keygen-helper

Generate and audit SSH keys for secure remote authentication.

context-manager

Manage persistent conversation context and session state with structured markdown files.

context-enhancer

Enhances user message intent and semantic clarity for AI response accuracy using Python 3.8+ templates.

ai-prompt-optimizer

Optimize natural language prompts for large language models with structural enhancements.

infinite-loop

Detect and mitigate infinite loop vulnerabilities in AI agent execution trajectories.

pentest-access-control

Identifies and validates web/API authorization vulnerabilities including RBAC bypass and privilege escalation.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install Agent3Sigma-Canary?▼

Run `npx skills add antgroup/Agent3Sigma-Canary --all -g -y` in your terminal to install all skills in this suite globally.

What does Agent3Sigma-Canary test?▼

It drives agents through realistic attack scenarios like prompt injection, memory poisoning, and malicious skills, then scores whether the agent stayed safe while still completing the task.

Which agent frameworks does it support?▼

It supports OpenClaw, Hermes, and NanoClaw, plus OpenClaw variants with defense plugins like Shield, SecureClaw, and ClawKeeper for comparing protection effectiveness.

Does it require Docker to run evaluations?▼

Yes. Each evaluation task runs in an isolated Docker container so high-risk attack samples cannot harm your host machine or other tasks.

Can I add my own evaluation tasks?▼

Yes. Task definitions and environments are modular, and the skill-to-sandbox workflow lets you convert real skills into custom sandboxed evaluation environments.