Back to skill

Security audit

AdsTurbo AI 视频生成 · AdsTurbo AI Video Generation

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because its upload instructions lack privacy warnings and consent checks before local media is sent to public links; scanning was also partial.

Consider the privacy implications before installing: the reported upload workflow may send selected local media to a third party and create publicly accessible links without first explaining this or asking for approval. The scan is incomplete: Skillspector inspected 11 of 12 components, reported malformed manifest data and unresolved references, and AIG completed without findings. The evidence therefore does not establish that the artifact is safe.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (10)

Tainted flow: 'request' from os.environ.get (line 122, credential/environment) → urllib.request.build_opener.open (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/shared/client.py (reported line 133)May include surrounding context.

python
},
        )
        try:
            with _OPENER.open(request, timeout=timeout) as resp:
                return resp.status, dict(resp.headers), resp.read().decode("utf-8", "replace")
        except urllib.error.HTTPError as err:
            return err.code, dict(err.headers or {}), err.read().decode("utf-8", "replace")

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 28)May include surrounding context.

md
e user the image at `captcha_url` (or the local file `captcha_file`), ask them to type the characters, then run `python3 scripts/login.py captcha --code <what they typed>`. The user must read and type the captcha themselves; never solve it on their behalf
3. Once it returns `"next": "enter_email_code"`, ask the user for the code AdsTurbo emailed them (check spam too) and run `python3 scripts/login.py verify --code <code>`
4. On success the API key is saved on this machine in `~/.config/adsturbo/credentials.json` and every script picks it up; just re-run the command that needed it

A new email is signed up automatically; an existing user is simply logged in and no duplicate key is created. `login.py status` shows whether a key is available; run `login.py logout` before switching accounts. The old way still works too: sign up at https://adsturbo.ai?channel=clawhub, create a key at https://adsturbo.ai/iris/api and `export ADSTURBO_API_KEY=<key>`; the environment variable takes precedence.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 124)May include surrounding context.

md
1. 问用户的邮箱,执行 `python3 scripts/login.py start --email <邮箱>`
2. 如果返回 `"next": "enter_captcha"`:把 `captcha_url` 这张图(或本地文件 `captcha_file`)给用户看,请他照图输入字符,再执行 `python3 scripts/login.py captcha --code <用户输入>`。图形验证码必须由用户本人辨认输入,不要自己识别代填
3. 返回 `"next": "enter_email_code"` 后,请用户去邮箱(也看看垃圾箱)找 AdsTurbo 发来的验证码,执行 `python3 scripts/login.py verify --code <验证码>`
4. 成功后 API Key 保存在本机 `~/.config/adsturbo/credentials.json`,之后所有脚本自动使用,直接重跑刚才的命令

新邮箱自动注册,老用户就是登录,不会重复创建 Key。`login.py status` 查看是否已登录,换账号先 `login.py logout`。也可以照旧在 https://adsturbo.ai?channel=clawhub 注册后到 https://adsturbo.ai/iris/api 创建 Key 并 `export ADSTURBO_API_KEY=<Key>`,环境变量优先。

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/login.py (reported line 10)May include surrounding context.

python
captcha --code    only when start said next=enter_captcha: what the user read off the image
  verify --code     the emailed code -> account (new or existing) -> API key saved locally

The key goes to ~/.config/adsturbo/credentials.json (0600), keyed by site, and
every other script picks it up when ADSTURBO_API_KEY is unset. It is never
printed: output only carries its last four characters.
"""

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/login.md (reported line 33)May include surrounding context.

md
## Where the key lives

- File: `~/.config/adsturbo/credentials.json` (mode 600), saved per site so the test and production sites never mix; set `ADSTURBO_CONFIG_DIR` to use another directory
- Every script reads it when `ADSTURBO_API_KEY` is unset; the environment variable wins
- If the account already has a key named `adsturbo-skill`, it is reused rather than duplicated
- Output only carries the key's last four characters; never echo the full key in a reply

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/login.md (reported line 85)May include surrounding context.

md
## Key 存在哪

- 文件:`~/.config/adsturbo/credentials.json`(权限 600),按站点分别保存,测试站和正式站互不影响;可用环境变量 `ADSTURBO_CONFIG_DIR` 换目录
- 所有脚本在 `ADSTURBO_API_KEY` 没设置时自动读它;环境变量优先
- 账号里已有名为 `adsturbo-skill` 的 Key 就直接复用,不会重复创建
- 输出只带 Key 的末 4 位,回复里不要回显完整 Key

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/shared/client.py (reported line 54)May include surrounding context.

python
"scripts/login.py (email + verification code), or set ADSTURBO_API_KEY."
)
CONFIG_DIR_ENV = "ADSTURBO_CONFIG_DIR"
CREDENTIALS_FILE = "credentials.json"


def config_dir() -> Path:

The local-media upload workflow lacks a privacy warning about third-party transfer and public accessibility.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow directs the agent to upload local media to obtain publicly accessible URLs without first explaining the privacy implications or obtaining informed consent. Although it states that public URLs are required, users requesting video editing may not realize their private source media will leave the machine and become accessible to anyone possessing the URL.

Content

No source excerpt is available for this finding.

The upload workflow lacks a privacy warning before publishing personal photos and voice recordings at publicly accessible URLs.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The example directs uploading a personal photograph and voice recording into a workflow that requires publicly accessible asset URLs, without a privacy warning or consent checkpoint. These identity-bearing assets could be accessed by unintended recipients if their URLs are exposed and could facilitate impersonation or unauthorized voice cloning.

Content

No source excerpt is available for this finding.

Local media uploads are instructed without a privacy warning.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The instructions automatically bridge local media into publicly accessible remote assets without requiring disclosure or approval before upload. In a video-generation workflow, users may supply private recordings or confidential product footage without intending to make the source material accessible through public links.

Content

No source excerpt is available for this finding.