Back to skill

Security audit

Wow Daily News

Security checks for vulnerabilities and agentic risk

Overview

This daily-news skill is mostly aligned with generating a report, but it uses high-impact automation with hard-coded message recipients, unverified local executables, a background service, and a command-injection-prone script.

Review before installing. This skill can create external documents and send links to fixed Feishu and WeChat accounts, run unverified local scripts and a native binary, and start a background local service. It should be limited to a trusted environment, with recipients made configurable, explicit confirmation before sending, bundled or pinned dependencies, and the shell=True command execution fixed before routine use.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/wow_daily_news.py:18
Finding

Remote Command Injection Through an Untrusted Article URL

Content
View full analysis
tuple[bool, str]: try: result = subprocess.run(cmd, shell=True, capture_output=True, text=True, timeout=timeout) return result.returncode == 0, result.stdout + result.stderr except Exception as e: return False, str(e) def close_browser(): run_cmd('agent-browser close', 10) def open_page(url: str) -> bool: run_cmd('agent-browser close', 10) time.sleep(1) success, _ = run_cmd(f'agent-browser open "{url}" --timeout 45000', 60) return success ``` The URL supplied to `open_page()` is derived from remotely retrieved EXWIND page content: ```python if '/url: /post/' in line and current.get('title'): match = re.search(r'/url: (/\S+)', line) if match: current['url'] = f"https://exwind.net{match.group(1)}" ``` ### Technical Analysis `run_cmd()` invokes `subprocess.run()` with `shell=True`. The `open_page()` function embeds its `url` argument directly into a shell command string. The URL is surrounded by double quotes, but quoting alone does not prevent command injection. An attacker-controlled value can contain a double quote followed by shell syntax, terminate the intended argument, and introduce another command. The parser accepts any non-whitespace characters after `/url:`, without validating the resulting URL's hostname, path, characters, or canonical form. Because article URLs are parsed from remote webpage content and later passed to `get_exwind_article_content()`, which invokes `open_page()`, compromised or attacker-influenced source content can reach the operating-system shell. ### Attack Path 1. An attacker compromises EXWIND content, its delivery path, or another mechanism capable of influencing the browser snapshot. 2. The attacker supplie ...[truncated 1245 chars]
Remediation
View remediation
tuple[bool, str]: try: result = subprocess.run( args, shell=False, capture_output=True, text=True, timeout=timeout, check=False, ) return result.returncode == 0, result.stdout + result.stderr except Exception as exc: return False, str(exc) def close_browser(): run_cmd(["agent-browser", "close"], 10) def open_page(url: str) -> bool: close_browser() time.sleep(1) success, _ = run_cmd( ["agent-browser", "open", url, "--timeout", "45000"], 60, ) return success ``` 2. Validate every remotely derived URL before use: ```python from urllib.parse import urlparse def validate_exwind_url(url: str) -> str: parsed = urlparse(url) if parsed.scheme != "https": raise ValueError("Unexpected URL scheme") if parsed.hostname != "exwind.net": raise ValueError("Unexpected URL host") if not parsed.path.startswith("/post/"): raise ValueError("Unexpected URL path") if parsed.username or parsed.password or parsed.port: raise ValueError("Unexpected URL authority") return url ``` 3. Reject control characters and malformed URLs rather than attempting to sanitize shell syntax. 4. Apply URL validation at the trust boundary, immediately after parsing remote content. 5. Run the Skill under a least-privileged account with restricted filesystem and network access as defense in depth. 6. Add regression tests containing quotes, command substitutions, semicolons, newlines, and other shell metacharacters. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:48
Finding

Execution of Unbundled and Unverified Local Components

Content
View full analysis
> mcp.log 2>&1 & sleep 10 && curl -s --connect-timeout 3 http://localhost:18060/mcp ``` ```bash python3 ~/.openclaw/workspace/skills/daily-beauty/daily_beauty.py ``` ### Technical Analysis The Skill directs the Agent to execute two Python scripts and one native binary that are not included in the audited project: - `~/.openclaw/workspace/scripts/daily_report_generator.py` - `~/xiaohongshu-mcp/xiaohongshu-mcp-linux-amd64` - `~/.openclaw/workspace/skills/daily-beauty/daily_beauty.py` The project does not define their trusted origin, pinned version, expected cryptographic digest, signature, ownership, or required filesystem permissions. Consequently, the behavior performed by the Skill can differ from the behavior represented by the reviewed artifact. The native MCP component is also launched through `nohup` as a detached background process. Although this does not establish cross-reboot persistence, it allows an unverified component to continue running after the immediate command or Skill step has completed. ### Attack Path 1. An attacker gains the ability to create or replace one of the referenced files before the Skill runs. Possible prerequisites include prior access to the Agent account, a writable shared workspace, a compromised installer, or an insecure update process. 2. The attacker places a malicious Python script or binary at an expected path. 3. A user invokes the daily-news Skill or its scheduled task runs. 4. The Agent follows the mandatory instructions ...[truncated 1036 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (15)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

该代码的核心功能是通过 agent-browser 打开 exwind.net,抓取魔兽世界相关新闻,筛选当天内容,获取详情页正文,格式化为 Markdown,并保存到本地 ~/.openclaw/workspace/img 目录下的 .md 文件。虽然与声明中的“魔兽世界新闻”部分一致,但声明强调的是一个完整的“每日日报自动生成与分发”流程,包含多个内容来源(魔兽世界新闻、NGA热帖、今日美图)、固定时间触发(18:00)、以及输出到飞书文档并推送到飞书和微信。上述关键能力在代码中均未体现,因此描述与实际行为存在明显不一致。

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill description does not warn users that execution sends a link to fixed Feishu and WeChat recipients. Hidden external delivery is dangerous because users may provide content under the assumption it stays local, while the skill actually performs outbound messaging to predetermined accounts.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
96% confidence
Finding

This is a true tool-parameter abuse issue because the code constructs shell commands dynamically and executes them through the shell. In the context of an automation skill that opens remote URLs and controls an external browser tool, this is more dangerous: untrusted content or future caller-controlled parameters could pivot into arbitrary local command execution.

Content

Scanner excerpt · scripts/wow_daily_news.py (reported line 20)May include surrounding context.

python
def run_cmd(cmd: str, timeout: int = 30) -> tuple[bool, str]:
    try:
        result = subprocess.run(cmd, shell=True, capture_output=True, text=True, timeout=timeout)
        return result.returncode == 0, result.stdout + result.stderr
    except Exception as e:
        return False, str(e)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill instructs the agent to use shell commands, write files, and manipulate local paths, but it does not declare any explicit tool scope or permission boundaries. This creates a hidden capability gap where a user may invoke a seemingly simple news skill that can execute host commands and modify local state without clear disclosure or enforcement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill description and all output instructions are written to generate a Chinese daily report, and there is no indication that users can opt into another language. This creates a locale/language restriction without user choice or an explicit documented justification that the skill is intentionally Chinese-only.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill collects third-party forum content and images, then inserts them into an external Feishu document without an upfront warning in the description. This creates privacy, copyright, and data-handling risks, especially if users do not expect third-party scraping and redistribution.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The invocation section does not clearly constrain when the skill should activate or what preconditions apply before it performs side effects. In context, that ambiguity is risky because the workflow culminates in sending a generated document to fixed Feishu and WeChat targets, making misfires materially harmful.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger phrase includes extremely generic language such as '日报', which can match ordinary conversation and cause the skill to run unexpectedly. Because the skill also creates documents and sends messages to fixed external recipients, accidental invocation can lead to unintended data collection, document creation, and outbound delivery.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
96% confidence
Finding

The use of 'nohup ... &' creates a background process that persists beyond the current session, which is unnecessary and risky for a daily news generation skill. Session persistence can be abused to maintain unauthorized services, consume resources, or provide a foothold for later misuse, especially when paired with hard-coded local executable paths.

Content

Scanner excerpt · SKILL.md (reported line 97)May include surrounding context.

text
   - 无响应 → 启动:
     ```bash
     cd ~/xiaohongshu-mcp && nohup ./xiaohongshu-mcp-linux-amd64 >> mcp.log 2>&1 & sleep 10 && curl -s --connect-timeout 3 http://localhost:18060/mcp
     ```

   **b. 重新运行美图脚本:**

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill directs the agent to start a local MCP service in the background using nohup, changing persistent host state beyond the immediate task of generating a daily report. Allowing a content-generation skill to launch local services increases attack surface, enables long-lived processes, and can be abused to establish persistence or run unintended binaries.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The module description and all user-facing output indicate the skill is specifically designed to produce Chinese-language news summaries, but there is no user opt-in or configurable language selection. This is a natural-language locale policy concern because the skill imposes a specific language across its interface and outputs.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
94% confidence
Finding

The helper wraps shell command execution with shell=True, which makes any future or indirect input passed into cmd subject to shell interpretation. In this file, open_page() interpolates a URL into the command string, so a malicious or malformed value containing shell metacharacters could trigger command injection and arbitrary command execution under the agent's privileges.

Content

Scanner excerpt · scripts/wow_daily_news.py (reported line 20)May include surrounding context.

python
def run_cmd(cmd: str, timeout: int = 30) -> tuple[bool, str]:
    try:
        result = subprocess.run(cmd, shell=True, capture_output=True, text=True, timeout=timeout)
        return result.returncode == 0, result.stdout + result.stderr
    except Exception as e:
        return False, str(e)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The stated skill scope includes three content sources/categories: 魔兽世界新闻, NGA 热帖, and 今日美图. The actual code only opens exwind.net, parses EXWIND article listings and article detail pages, and never retrieves NGA posts or any image content.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes an end-to-end daily report workflow including automatic generation of a Feishu document and delivery to Feishu and WeChat. In this file, the implemented behavior stops at scraping EXWIND content, formatting Markdown, and saving it locally; there is no Feishu API call, WeChat push, or document creation logic.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The skill's stated purpose is to compile and send a daily report. Referring to host service management via systemctl indicates infrastructure inspection capability, which is not part of the manifest's declared user-facing function and is only tangentially related to fetching beauty images.

Content

No source excerpt is available for this finding.