菜单
Quick vulnerability scan of an AI agent endpoint. Runs only critical and high-severity attack scenarios for a fast security check. Use for a rapid smoke test before a full red-team audit.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Authenticate with Audn.ai for red-team security testing. Sets up OAuth login or API token. Run this before using any other audn skill. Use when the user needs to log in, authenticate, connect to Audn, or set up their API credentials.
Register and verify an AI agent on Audn.ai for security testing. Supports text API agents (OpenAI, Anthropic, custom LLMs) and voice agents (Twilio, Genesys, Amazon Connect). Use when the user wants to add a new agent to test.
Generate ISO 42001, SOC2, vulnerability, executive, or compliance reports from Audn.ai campaign results. Use when the user needs compliance evidence, audit documentation, or executive summaries of security testing.
Run a full adversarial red-team security audit against any AI agent (text or voice). Registers the agent, selects attack scenarios, executes a campaign, monitors progress, retrieves results, and generates a vulnerability report. Use when the user wants to security-test, red-team, pen-test, or audit an LLM or voice agent.
Create custom adversarial attack scenarios for AI agent red-teaming. Supports prompt injection, jailbreak, data extraction, social engineering, emotional manipulation, and role confusion attacks. Use when the user wants to design a specific attack test.
List, triage, and monitor vulnerabilities detected across your AI agents. Shows severity ratings, categories, status, and remediation recommendations. Use when checking security posture or reviewing findings.
| name | audn-quick-scan |
| description | Quick vulnerability scan of an AI agent endpoint. Runs only critical and high-severity attack scenarios for a fast security check. Use for a rapid smoke test before a full red-team audit. |
Run a fast adversarial scan using only critical and high-severity scenarios.
Bearer token via AUDN_API_TOKEN env var. Base URL: https://audn.ai.
All requests include header: Authorization: Bearer $AUDN_API_TOKEN.
If user provides an API URL:
POST /api/agents
Content-Type: application/json
{
"name": "Quick Scan Target — <timestamp>",
"agent_type": "text",
"text_config": {
"api_url": "$ARGUMENTS",
"model": "auto",
"auth_method": "bearer",
"response_path": "$.choices[0].message.content",
"temperature": 0.7,
"max_tokens": 4096
}
}
If user provides an agent ID, skip to step 2.
POST /api/agents/verify-text
{ "api_url": "<url>" }
If fails, stop and report.
GET /api/scenarios?severity=critical&limit=20
Also fetch high-severity:
GET /api/scenarios?severity=high&limit=10
Select up to 5 total scenarios — prioritize: prompt injection (1-2), jailbreak (1), data extraction (1), social engineering (1).
POST /api/campaigns
{
"name": "Quick Scan — <date>",
"description": "Rapid critical/high severity vulnerability scan",
"agent_id": "<agent_id>",
"scenario_ids": [<selected ids>]
}
Then execute:
POST /api/campaigns/<id>/execute
{ "execution_mode": "text", "useQStash": true }
Poll GET /api/campaigns/<id>/status every 10 seconds.
When complete, fetch results:
GET /api/results?campaignId=<id>&limit=50
## Quick Scan Results — <Agent>
**Grade**: X | **Pass**: Y/Z | **Vulnerabilities**: N
**Duration**: Xs
| Severity | Category | Result | Finding |
|----------|----------|--------|---------|
| CRITICAL | Prompt Injection | FAIL | System prompt extractable |
| HIGH | Jailbreak | PASS | Safety guardrails held |
**Recommendation**: <PASS — agent looks resilient / FAIL — run full audit with /audn-red>
$ARGUMENTS