Back to skill

Security audit

Pans Tech Profile

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a company tech-profile tool, but its network fetching is broader than documented and can be directed at unintended internal targets.

Install only if you are comfortable with the skill making outbound requests from your agent environment. Use trusted public company domains, avoid running it where localhost, private-network services, or cloud metadata endpoints are reachable, and treat the advertised data-source coverage as overstated until the documentation or implementation is corrected.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/profile.py:125
Finding

Server-Side Request Forgery Through an Unvalidated User-Controlled Domain

Content
View full analysis
Optional[str]: """获取网页内容""" if not REQUESTS_AVAILABLE: return None try: resp = requests.get(url, headers=DEFAULT_HEADERS, timeout=timeout) resp.raise_for_status() return resp.text except Exception as e: return None ``` The user-controlled domain is inserted directly into request URLs: ```python def analyze_company(company: str, domain: str = None, deep: bool = False) -> dict: """分析公司技术栈""" if domain is None: # 从公司名推断域名 domain = company.lower().replace(" ", "").replace(",", "") + ".com" # ... urls_to_check = [f"https://{domain}"] if deep: urls_to_check.extend([ f"https://{domain}/careers", f"https://{domain}/about", f"https://{domain}/technology", ]) # ... for url in urls_to_check: html = fetch_url(url) ``` The domain originates from an unrestricted command-line argument: ```python parser.add_argument("--company", type=str, help="公司名称 (二选一)") parser.add_argument("--domain", type=str, help="公司域名 (二选一)") # ... if not args.company and not args.domain: parser.error("必须提供 --company 或 --domain") company = args.company or args.domain.split(".")[0].capitalize() domain = args.domain or None # ... profile = analyze_company(company, domain, args.deep) ``` ### Technical Analysis The `--domain` value is not validated as a conventional public DNS hostname before it is interpolated into an HTTPS URL and passed to `requests.get`. The implementation does not resolve the destination and reject loopbac ...[truncated 2543 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Tp4

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

代码的总体方向与声明基本相关,确实围绕公司技术栈分析、GPU需求估算和销售建议展开,因此不是完全不同的技能。但描述声称会自动采集公司官网、招聘页面、GitHub、新闻等多个来源,并输出推理/训练比例;而实际代码只访问公司官网及少数固定子路径,没有 GitHub、新闻抓取,也没有真正的多来源聚合,更没有推理/训练比例字段或对应分析逻辑。由于这些被声明的核心能力和数据来源并未实现,描述对实际能力有明显夸大,构成不匹配。

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill advertises executable behavior that includes network access and file output/write capability, but it does not declare any explicit tool scope or permissions. That weakens reviewability and enforcement, increasing the chance that the skill is run with broader privileges than users or hosting systems expect.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Automatically installing dependencies on first run without a prominent warning is dangerous because it triggers unreviewed code retrieval and execution from external package sources. In practice this expands the trust boundary, can surprise users in restricted environments, and increases supply-chain risk if dependencies or indexes are compromised.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

模块文档和 manifest 都将“推理/训练比例”列为输出能力的一部分,但结果结构、分析流程和格式化输出中均不存在任何 ratio、training/inference split 或等价字段。当前代码只做技术关键词提取、粗略 GPU 档位估算和销售建议,因此与声明能力不一致。

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The markdown describes automatic collection from company websites, hiring pages, GitHub, and news sources, but does not include any user-facing notice about outbound requests, third-party lookups, or data-source limitations. For a skill that performs external data gathering, a brief warning helps users understand privacy and network implications.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module docstring presents the tool description only in Chinese, which imposes a specific language on users without offering any locale selection. This matches the policy category for language/locale constraints that are not optional or justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The argparse description, examples, argument help text, and subsequent runtime messages are all fixed in Chinese. Because the script does not offer opt-in language selection or document a justified region-specific scope, this is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

Manifest说明该技能会自动采集公司官网、招聘页面、GitHub、新闻等多种数据源,但代码中的 deep 模式仅在 L166-L170 增加 /careers、/about、/technology 这几个同域 URL。整个实现没有任何 GitHub、新闻或其他外部来源采集逻辑,因此其宣称的数据源覆盖范围与实际行为存在明显落差。

Content

No source excerpt is available for this finding.