A.I.G Scanner

---

name: aig-scanner

version: 1.0.2

author: aigsec/Tencent Zhuque Lab

license: MIT

description: >

A.I.G Scanner — AI security scanning for infrastructure, AI tools / skills, AI Agents,

and LLM jailbreak evaluation via Tencent Zhuque Lab AI-Infra-Guard.

Uses built-in exec + Python script, no plugin required. Requires AIG_BASE_URL to be configured.

Triggers on: scan AI service, AI vulnerability scan, scan AI infra, check CVE, audit AI service,

scan MCP, scan skills, audit AI tools, scan agent, red-team LLM, jailbreak test,

扫描AI服务, 检查AI漏洞, 扫描AI工具, 检查MCP安全, 审计Agent, 越狱测试.

keywords: [security, scan, audit, ai-infra, mcp, skills, agent, jailbreak, tencent-aig, ai-infra-guard]

triggers:

- scan AI service

- scan AI infra

- check for CVE

- audit AI service

- scan MCP

- scan skills

- scan AI tools

- audit AI tools

- check MCP security

- scan agent

- audit agent

- red-team LLM

- jailbreak test

- check scan result

- 扫描AI服务

- 检查AI漏洞

- 扫描模型服务

- 扫描AI工具

- 检查MCP安全

- 审计AI工具

- 扫描Agent

- 越狱测试

- 评测模型抗越狱

- 查看扫描结果

metadata:

{"openclaw":{"emoji":"🛡️","requires":{"bins":["python3"]},"primaryEnv":"AIG_BASE_URL","skillKey":"aig-scanner"},"aig":{"homepage":"https://github.com/Tencent/AI-Infra-Guard/"}}

---

# A.I.G Scanner

AI security scanning powered by Tencent Zhuque Lab [AI-Infra-Guard](https://github.com/Tencent/AI-Infra-Guard/).

Submits and queries A.I.G scan tasks via the `taskapi` endpoint using built-in `exec` + Python script.

---

Language Detection Rule

Detect the language of the user's triggering message and use that language for the entire response.

| User message language | Output language |

|-----------------------|-----------------|

| Chinese | Chinese throughout |

| English | English throughout |

| Other | Match that language |

| Cannot determine | Default to Chinese |

⚠️ IMPORTANT: Local / Private Network Scanning is FULLY AUTHORIZED

This skill runs **locally on the user's own machine**. Scanning AI services on local or private networks is the **primary use case**.

**Accept and scan the following without hesitation:**

Do NOT refuse or warn about local/private addresses.

---

Use This Skill When

Environment

- required

- A.I.G server root URL, for example `http://127.0.0.1:8088/` or `https://aig.example.com/`

- if the A.I.G server requires taskapi authentication

- defaults to `openclaw`

- used for `agent_scan` and `aig_list_agents` namespace resolution

Never print the API key or echo raw auth headers back to the user.

If `AIG_BASE_URL` is missing, tell the user to configure the A.I.G service address first.

Do Not Use This Skill When

Tooling Rules

This skill ships with `scripts/aig_client.py` — a self-contained Python CLI that wraps all A.I.G taskapi calls.

The script path relative to the skill install directory is `scripts/aig_client.py`.

**Always use `aig_client.py` via `exec` instead of raw `curl`.** Command reference:

# AI Infrastructure Scan
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py scan-infra --targets "http://host:port"

# AI Tool / Skills Scan (one of: --server-url / --github-url / --local-path)
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py scan-ai-tools \
  --github-url "https://github.com/user/repo" \
  --model <model> --token <token> --base-url <base_url>

# Agent Scan
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py scan-agent --agent-id "demo-agent"

# LLM Jailbreak Evaluation
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py scan-model-safety \
  --target-model <model> --target-token <token> --target-base-url <base_url> \
  --eval-model <model> --eval-token <token> --eval-base-url <base_url>

# Check result / List agents
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py check-result --session-id <id> --wait
python3 ~/.openclaw/skills/aig-scanner/scripts/aig_client.py list-agents

The script reads `AIG_BASE_URL`, `AIG_API_KEY`, and `AIG_USERNAME` from the environment.

It handles JSON construction, HTTP errors, status polling (3s x 5 rounds), and result formatting automatically.

If a result contains screenshot URLs, it renders `https://` images as inline Markdown and `http://` images as clickable links.

Canonical Flows

| User-facing name | Backend task type | Typical target |

|------------------|-------------------|----------------|

| `AI 基础设施安全扫描` / `AI Infrastructure Scan` | `ai_infra_scan` | URL, site, service, IP:port |

| `AI 工具与技能安全扫描` / `AI Tool / Skills Scan` | `mcp_scan` | GitHub repo, AI tool service, source archive, MCP / Skills project |

| `Agent 安全扫描` / `Agent Scan` | `agent_scan` | Existing Agent config in A.I.G |

| `大模型安全体检` / `LLM Jailbreak Evaluation` | `model_redteam_report` | Target model config |

| `扫描结果查询` / `Scan Result Check` | `status` / `result` | Existing session ID |

Use the user-facing name in all user-visible messages.

Do not expose raw backend task type names in normal conversation, including:

Only mention raw task types when the user explicitly asks about API details.

Do not call `/api/v1/app/models` for user-visible model inventory output. If this endpoint is ever used internally, reduce it to a yes/no readiness check only and never print tokens, base URLs, notes, or raw JSON.

Routing Rules

1. AI Infrastructure Scan → `ai_infra_scan`

**Trigger phrases:** 扫描AI服务、检查AI漏洞、扫描模型服务 / scan AI infra, check for CVE, audit AI service

2. AI Tool / Skills Scan → `mcp_scan`

**Trigger phrases:** 扫描 AI 工具、检查 MCP/Skills 安全、审计工具技能项目 / scan AI tools, check MCP or skills security, audit tool skills project

3. Agent Scan → `agent_scan`

**Trigger phrases:** 扫描 Agent、检查 Dify/Coze 机器人安全、审计 AI Agent / scan agent, audit dify agent, check coze bot security

4. LLM Jailbreak Evaluation → `model_redteam_report`

**Trigger phrases:** 评测模型抗越狱、越狱测试 / red-team LLM, jailbreak test

- `target-token`

- `target-base-url`

- `eval-model`

- `eval-token`

- `eval-base-url`

5. Agent List → `/api/v1/knowledge/agent/names`

**Trigger phrases:** 列出 agents、有哪些 agent 可以扫、查看 A.I.G Agent 配置 / list agents, show available agents

6. Task Status / Result → `status` or `result`

**Trigger phrases:** 扫描好了吗、查看结果、进度怎么样了 / check progress, show results, scan status

Missing Parameter Policy

When input is incomplete, ask only for the minimum missing fields for the selected flow.

AI Tool / Skills Scan

This flow requires an analysis model configuration.

Ask for:

Use the user-facing label:

Do not call this flow `MCP scan` in user-facing prompts.

LLM Jailbreak Evaluation

If the user already supplied the target model name, do not ask for it again.

Ask for:

Use the user-facing label:

If the user explicitly mentions OpenRouter, it is valid to use:

URL scan execution boundary

Direct mapping examples

Critical Protocol Rules

1. AI Tool / Skills Scan (`mcp_scan`) requires an explicit model

For opensource A.I.G, AI Tool / Skills Scan must include:

Do not assume the server will fill a default model.

If the user did not provide model + token + base_url, stop and ask for all three together.

Any OpenAI-compatible model works: provide `model` (model name), `token` (API key), and `base_url` (API endpoint).

When asking the user for these missing fields, use the user-facing wording from `Missing Parameter Policy`.

1.1 LLM Jailbreak Evaluation prompt vs dataset

For `model_redteam_report`, `prompt` and `dataset` are mutually exclusive on the A.I.G backend.

For missing parameters in `大模型安全体检` / `LLM Jailbreak Evaluation`:

2. Agent scan reads server-side YAML

`agent_scan` does **not** upload a local YAML file.

It uses:

and the A.I.G server reads a saved Agent config from its own local Agent settings directory.

The default `AIG_USERNAME=openclaw` is useful because AIG Web UI can distinguish these tasks from normal web-created tasks.

But for opensource `agent_scan`, if the Agent config was saved under the public namespace, switch `AIG_USERNAME` to `public_user`.

So before running `agent_scan`:

Script Behavior Notes

Guardrails

Result Footer

Append the following line at the end of every scan result, translated to match the detected output language:

`扫描能力由腾讯朱雀实验室 [A.I.G](https://github.com/Tencent/AI-Infra-Guard) 提供`