Red-team your AI agents from any coding IDE. Automated adversarial security testing for voice and text AI agents — jailbreaks, prompt injection, data extraction, social engineering, compliance reporting.
Works with Claude Code, Cursor, Codex, Windsurf, Cline, Amp, and 40+ other coding agents.
Audn.ai account — sign up free
npx skills add audn-ai/skills --all -a claude-codeIn Claude Code, run:
/audn-login
This registers the Audn.ai MCP server. Then:
- Restart Claude Code — exit with
/exitorCtrl+C, then runclaudeagain (MCP servers load at startup) - Run
/audn-loginagain - Run
/mcp→ selectaudn-redteam→ click "Authenticate" → log in via browser
After login, tokens are stored in your system keychain and refresh automatically. No API keys needed.
Not using Claude Code? Generate a token at audn.ai/dashboard/settings, then:
export AUDN_API_TOKEN="your-bearer-token"/audn-red https://api.openai.com/v1/chat/completions
That's it. Audn registers your agent, selects attack scenarios, runs the campaign, and delivers a full vulnerability report.
# Single skill
npx skills add audn-ai/skills --skill audn-red -a claude-code
# All skills
npx skills add audn-ai/skills --all -a claude-code
# Multiple agents at once
npx skills add audn-ai/skills --skill audn-red -a claude-code -a cursor -a codex
# Global install (available in all projects)
npx skills add audn-ai/skills --skill audn-red -a claude-code -g
# Claude Code plugin mode
claude --plugin-dir ./ # from this repoSet up authentication. On Claude Code, triggers OAuth browser login via the bundled MCP server. On other agents, guides API token setup.
/audn-login
End-to-end adversarial security campaign. Registers your agent, verifies connectivity, selects attack scenarios from the library, creates and executes a campaign, monitors progress, retrieves results, checks for vulnerabilities, and generates a report.
/audn-red https://api.openai.com/v1/chat/completions
/audn-red https://api.anthropic.com/v1/messages
/audn-red https://your-internal-api.com/chat
Output: Security grade (A-F), vulnerability table, attack category breakdown, remediation recommendations.
Runs only critical and high-severity scenarios for a rapid smoke test. Takes minutes instead of a full audit.
/audn-quick-scan https://api.openai.com/v1/chat/completions
Generate ISO 42001, SOC2, vulnerability, executive, or general compliance reports from completed campaign data.
/audn-compliance-report <campaign-id> iso_42001
/audn-compliance-report <campaign-id> soc2
/audn-compliance-report <campaign-id> executive
Register text API agents (OpenAI, Anthropic, custom LLMs) or voice agents (Twilio, Genesys, Amazon Connect) for testing.
/audn-agent-setup https://api.openai.com/v1/chat/completions
/audn-agent-setup +14155551234
Design custom adversarial attack scenarios — prompt injection, jailbreak, data extraction, social engineering, emotional manipulation, role confusion.
/audn-scenario-builder
List and triage vulnerabilities across all your agents. Severity ratings, categories, remediation recommendations, trend analysis.
/audn-vuln-monitor
| Attack Category | What It Tests |
|---|---|
| Prompt Injection | System prompt override, instruction hijacking, context manipulation |
| Jailbreak | Safety guardrail bypass, persona override, DAN-style attacks |
| Data Extraction | PII leakage, credential exposure, system prompt extraction |
| Social Engineering | Authority impersonation, urgency manipulation, trust exploitation |
| Emotional Manipulation | Sympathy exploitation, guilt tripping, rapport abuse |
| Role Confusion | Identity confusion, role-play exploitation, context switching |
These skills orchestrate the Audn.ai API:
| Resource | Operations |
|---|---|
| Agents | Create, list, delete, verify (text + phone) |
| Campaigns | Create, execute (voice + text), status, delete |
| Scenarios | Library browse, custom create, update, delete, execute |
| Results | List with filters, detail view, stats |
| Reports | Generate (ISO 42001, SOC2, vulnerability, executive, compliance) |
| Vulnerabilities | List with severity/category/status |
| API Keys | Create, list, revoke, rotate |
This repo doubles as a Claude Code plugin with a bundled MCP server that handles OAuth login automatically:
# Load directly (includes MCP server with OAuth)
claude --plugin-dir /path/to/this/repo
# Or add as marketplace
/plugin marketplace add audn-ai/skills
# Or install from official marketplace (after approval)
/plugin install audn-redteam@claude-plugins-officialWhen loaded as a plugin, the MCP server at mcp.audn.ai connects via OAuth 2.1 with PKCE and Dynamic Client Registration — no API keys needed. Run /mcp to authenticate via browser, then all 34 Audn API tools are available natively.
# Or add the MCP server standalone (without plugin)
claude mcp add --transport http audn-redteam https://mcp.audn.ai/mcp/audn-rn8sxThese skills are instruction files (SKILL.md) that teach your coding agent how to orchestrate the Audn.ai API. They contain zero proprietary code — just structured prompts that call the public API endpoints.
You (IDE) → Coding Agent → Skill Instructions → Audn.ai API → Results
The Audn.ai platform does the heavy lifting: adversarial prompt generation, multi-turn attack execution, vulnerability analysis, and compliance report generation.
Found a bug or want to improve a skill? PRs welcome.
- Fork this repo
- Edit skills in
skills/<skill-name>/SKILL.md - Test locally:
npx skills add ./ --skill <name> -a claude-code -y - Submit a PR
MIT — see LICENSE. Skills are open-source. The Audn.ai API is subject to Audn.ai Terms of Service.
audn.ai — Continuous security testing for AI agents.