Back to skill

Security audit

Free-Model Config

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because it automatically uploads local files to public hosts and provides inconsistent protection for API keys.

Use non-sensitive media until public uploads require clear approval of the files and destination, and inspect storyboard paths before running them. Keep API keys out of reports and protect configuration files and backups. Preview configuration changes before --write, and replace predictable temporary files. Explicitly choose the video model: the scripts default to one the documentation labels paid. Hosting retention, live pricing, and third-party dependency behavior were not independently verified.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (32)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The main generation and concatenation workflows match the declared Agnes AI functionality. However, upload_image_to_url sends local image bytes to external hosting services, rather than only to the selected AI provider, without a separate confirmation step. This introduces an undeclared external data recipient and is materially relevant to privacy. The supplied execution path also performs no official-model preflight check despite the description stating that this is mandatory. Other advertised provider and audio capabilities being absent from this single script is not independently treated as a mismatch.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 38)May include surrounding context.

md
| 智谱 BigModel | https://open.bigmodel.cn/api/paas/v4 | `glm-4.7-flash`(推荐,免费文本)、`glm-4.6v-flash`(免费视觉)、`cogview-3-flash`(免费图像)、`cogvideox-flash`(免费视频)、`glm-5.3-flash`(付费升级) | **部分免费**:仅免费档小模型,旗舰/OCR/语音付费 | 文本/视觉/图像/视频全覆盖 |
| 商汤 SenseNova | https://token.sensenova.cn/v1 | `sensenova-6.8-flash`(推荐)、`sensenova-6.8-flash-lite`、`sensenova-u1.5-fast`、`deepseek-flash`(V4.1)、`glm-5.2`、`kimi-k3` | 积分制(5h 12 万 / 周 120 万),旗舰可用 | 非"按模型免费",按积分扣减 |
| 小红书 dots studio | https://note3-prev-api.askdiandian.com/v1 | `dots3-note-prev`(推荐)、`dots_webdev_1022`(端点自报,未验证) | 预览期限时免费 | 认证头是 `api-key:`,**不是 Bearer** |
| 美团 LongCat | https://api.longcat.chat/openai | `LongCat-2.5`(推荐) | 免费额度 | 原生多模态,1M 上下文 |
| 小米 MiMo | https://token-plan-cn.xiaomimimo.com/v1 | `mimo-v2.6-pro`(推荐)、`mimo-v2.6-flash`、`mimo-v2.5-asr`、`mimo-v2.5-tts`(含 clone/design) | 新用户额度 + Token Plan(免费持续性最弱) | 旧版 ID 已停用,按上表配置 |

**排序规则**:平台按「地域 → 免费覆盖面 → 免费性质 → 免费额度 → 免费档质量」排序;平台内部按「文本 → 视觉 → 图像 → 视频 → 语音」排序。完整规则见下文「排序与分类逻辑」,增删模型/平台时必须同步全部清单。

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF2-模型选择域范本.md (reported line 15)May include surrounding context.

md
| 智谱 BigModel | glm-4.7-flash(推荐,部分免费)/ glm-5.3-flash(付费) | 文本、视觉、图像、视频 | https://open.bigmodel.cn/api/paas/v4 |
| 商汤 SenseNova | sensenova-6.8-flash(推荐)/ u1.5-fast / deepseek-flash / glm-5.2 / kimi-k3 | 多模态对话 / 图像 / 推理 | https://token.sensenova.cn/v1 |
| 小红书 dots studio | dots3-note-prev(推荐) | 长上下文 / 多模态 / 工具调用 | https://note3-prev-api.askdiandian.com/v1 |
| 美团 LongCat | LongCat-2.5(推荐) | 文本 / 视觉 / Agentic 编程 | https://api.longcat.chat/openai |
| 小米 MIMO | mimo-v2.6-pro(推荐)/ mimo-v2.6-flash / mimo-v2.5-asr / mimo-v2.5-tts 系列 | 文本 / 全模态 / 语音识别 / 语音合成 | https://token-plan-cn.xiaomimimo.com/v1 |

2. **平台能力对比**:

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF2-模型选择域范本.md (reported line 89)May include surrounding context.

md
| 智谱 BigModel | glm-4.7-flash(推荐) | 文本 / 视觉 / 图像 / 视频(部分免费) | https://open.bigmodel.cn/api/paas/v4 |
| 商汤 SenseNova | sensenova-6.8-flash(推荐) | 多模态对话 / 图像 / 推理 | https://token.sensenova.cn/v1 |
| 小红书 dots | dots3-note-prev(推荐) | 长上下文 / 多模态 / 工具调用 | https://note3-prev-api.askdiandian.com/v1 |
| 美团 LongCat | LongCat-2.5(推荐) | 文本 / 视觉 / Agentic 编程 | https://api.longcat.chat/openai |
| 小米 MIMO | mimo-v2.6-pro(推荐) | 文本 / 全模态 / 语音 | https://token-plan-cn.xiaomimimo.com/v1 |

## 平台能力对比

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 117)May include surrounding context.

md
| 智谱 BigModel | https://open.bigmodel.cn/api/paas/v4 | 文本生成接口 |
| 商汤 SenseNova | https://token.sensenova.cn/v1 | 文本生成接口 |
| 小红书 dots | https://note3-prev-api.askdiandian.com/v1 | OpenAI 兼容(认证头 api-key) |
| 美团 LongCat | https://api.longcat.chat/openai | 文本生成接口 |
| 小米 MIMO | https://token-plan-cn.xiaomimimo.com/v1 | 文本/语音接口 |

2. **接口参数**:

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 445)May include surrounding context.

md
"id": "LongCat-2.5",
    "name": "LongCat-2.5",
    "vendor": "Custom",
    "url": "https://api.longcat.chat/openai",
    "apiKey": "your_api_key_here",
    "supportsToolCall": true,
    "supportsImages": true,

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 627)May include surrounding context.

md
| 商汤 SenseNova | sensenova-u1.5-fast | POST https://token.sensenova.cn/v1/images/generations | 图像生成/编辑 |
| 商汤 SenseNova | deepseek-flash / glm-5.2 / kimi-k3 | POST https://token.sensenova.cn/v1/chat/completions | 对话推理 |
| 小红书 dots | dots3-note-prev | POST https://note3-prev-api.askdiandian.com/v1/chat/completions | 长上下文对话(头 `api-key`) |
| 美团 LongCat | LongCat-2.5 | POST https://api.longcat.chat/openai/chat/completions | 文本生成 |
| 小米 MIMO | mimo-v2.6-pro / mimo-v2.6-flash | POST https://token-plan-cn.xiaomimimo.com/v1/chat/completions | 文本/全模态 |
| 小米 MIMO | mimo-v2.5-asr | POST https://token-plan-cn.xiaomimimo.com/v1/audio/transcriptions | 语音识别 |
| 小米 MIMO | mimo-v2.5-tts / -voiceclone / -voicedesign | POST https://token-plan-cn.xiaomimimo.com/v1/audio/speech | 语音合成 |

The configuration-writing workflow persists API keys and creates backups without warning about credential exposure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The documented workflow propagates credential-bearing configuration into target files and automatically creates backups, without specifying secret-safe storage or backup handling. This can leave additional recoverable API-key copies accessible through permissive filesystem settings, synchronization, or later sharing, although actual file permissions cannot be verified without the script.

Content

No source excerpt is available for this finding.

A content-permissions clause imposes unconditional compliance with requests to modify the skill.

Medium
Category
Prompt Injection
Confidence
96% confidence
Finding

The clause forbids refusing any modification to the skill, without preserving authorization or safety boundaries. In an agent with write access, this can encourage accepting changes that remove safeguards or introduce unsafe behavior, though it cannot legitimately override higher-priority instructions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 155)May include surrounding context.

接口测试:

bash
curl -X POST {接口地址} \
  -H "Authorization: Bearer {API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 247)May include surrounding context.

配置测试:

bash
curl -X POST {接口地址} \
  -H "Authorization: Bearer {API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 571)May include surrounding context.

bash
## 测试 Agnes AI 模型
curl -X POST https://apihub.agnes-ai.com/v1/chat/completions \
  -H "Authorization: Bearer your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF3-配置生成域范本.md (reported line 696)May include surrounding context.

Agnes AI 图像生成:

bash
curl -X POST https://apihub.agnes-ai.com/v1/images/generations \
  -H "Authorization: Bearer {API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{

API-key validation instructs an external API call without warning that credentials will be transmitted to the configured endpoint.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The validation workflow explicitly instructs calling the API to test the configured key, rather than merely listing an available tool. It does not warn that this sends the credential to the configured service or require checking that the endpoint is trusted before transmitting it.

Content

No source excerpt is available for this finding.

The update-record template requests old and new API keys without a credential-exposure warning or explicit redaction requirement.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The update template requests old and new API-key values in a Markdown record without requiring redaction. An agent filling these fields could expose working credentials in conversations, reports, or retained logs; rotating a key does not protect the newly disclosed replacement.

Content

No source excerpt is available for this finding.

The update-report template calls for including both old and new API keys in Markdown output.

Medium
Category
Data Exfiltration
Confidence
97% confidence
Finding

The Markdown update-record fields explicitly invite inclusion of both previous and replacement API keys. Although the later example uses abbreviated dummy values, it does not impose redaction on actual reports, allowing credentials to escape their configuration storage boundary.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF5-使用指导域范本.md (reported line 32)May include surrounding context.

文本生成示例:

bash
curl -X POST https://apihub.agnes-ai.com/v1/chat/completions \
  -H "Authorization: Bearer your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/exemplars/MF5-使用指导域范本.md (reported line 213)May include surrounding context.

文本生成

bash
curl -X POST https://apihub.agnes-ai.com/v1/chat/completions \
  -H "Authorization: Bearer your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{

The generation workflow lacks a privacy warning about sending prompts and reference images to an external service.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow instructs the agent to generate images and submit a reference image for video generation through Agnes. Although the notes mention an API and publicly accessible image URLs, the file never warns that supplied prompts and reference imagery are shared with an external service or that sensitive images should not be made publicly accessible.

Content

No source excerpt is available for this finding.

The configuration workflow writes to a target configuration file without warning about effects on existing configuration.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

MF3-04 explicitly includes writing generated env and JSON configuration into a target file as part of the skill's own workflow, rather than merely listing another tool's capabilities. The file provides no warning that this can change existing settings; the generic reminder task at L59 does not disclose this risk.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fmc-requirements.md (reported line 19)May include surrounding context.

md
| 智谱 BigModel | https://docs.bigmodel.cn/cn/guide/start/model-overview | https://open.bigmodel.cn/api/paas/v4/models | https://docs.bigmodel.cn/cn/guide/start/pricing |
| 商汤 SenseNova | https://platform.sensenova.cn/docs | https://token.sensenova.cn/v1/models | 平台控制台账单页 |
| 小红书 dots | https://dots.ai/platform/apikeys | https://note3-prev-api.askdiandian.com/v1/models | 平台公告 |
| 美团 LongCat | https://longcat.chat/platform/docs | https://api.longcat.chat/openai/v1/models | 平台控制台 |
| 小米 MiMo | https://mimo.mi.com/docs/zh-CN/quick-start/summary/model | https://token-plan-cn.xiaomimimo.com/v1/models | https://mimo.mi.com/docs/zh-CN/updates/deprecate |

> 端点不带鉴权时可能返回 401:属正常,改用官方模型页核对,不要据此判定模型不存在。

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fmc-requirements.md (reported line 541)May include surrounding context.

md
##### 美团——LongCat
- 后台网址:https://longcat.chat/platform/api_keys
- 接口地址:https://api.longcat.chat/openai(OpenAI 兼容)
- API Key:从后台网址生成
- 模型名称:LongCat-2.5(推荐,1M 上下文 / 256K 输出,原生多模态)
- 能力:文本生成、视觉理解、Agentic 编程

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fmc-requirements.md (reported line 34)May include surrounding context.

curl -s https://apihub.agnes-ai.com/v1/models -H "Authorization: Bearer $KEY" | head -c 500

3) 单模型存活探测(最可靠)

curl -s -o /dev/null -w "%{http_code}\n" https://open.bigmodel.cn/api/paas/v4/chat/completions
-H "Authorization: Bearer $KEY" -H "Content-Type: application/json"
-d '{"model":"glm-4.7-flash","messages":[{"role":"user","content":"ping"}],"max_tokens":1}'

text

Image-upload instructions omit a warning about exposing user media to third-party hosting services.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The guidance recommends uploading images to third-party hosting services or retrying with other hosts without explaining the additional recipients or public-link exposure. An agent applying this workflow to private user images could disclose them beyond the intended model provider without informed approval.

Content

No source excerpt is available for this finding.