Back to skill

Security audit

Pub Gbb1010

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious because scanners report a gateway authentication secret in the package and concerns about cross-session persistence and external sharing of research terms; malicious intent was not established.

Review the credential exposure and privacy concerns before installing. The verdict is suspicious, not malicious. Skillspector inspected 7 of 8 components, with partial coverage for one component and a degraded static analyzer, so the evidence is incomplete.

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
config.json:3
Finding

Gateway authentication secret distributed in the skill package

Content
View full analysis

Impact

The confidentiality of the gateway authentication credential is lost, enabling direct requests under that credential wherever it remains accepted. The documented email-selected account workflow also creates account-isolation concerns, but cross-account access and token forgery were not verified and are not asserted as confirmed impacts.

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

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 388)May include surrounding context.

md
| **单次查询条数**| PageSize 20-1000,建议 ≤100 以保证速度|
| **翻页上限**| 无硬限制,但建议不超过 50 页(共 1000 条)/ No hard limit, but ≤50 pages recommended |
| **请求频率**| 有频率限制(未公开数值),触发 429 时自动等待重试|
| **Token 有效期**| Access Token 30 秒,Refresh Token 2 小时(以下游 API 返回为准)。系统自动管理刷新|
| **下载链接有效期**| 约 10 分钟(以下游返回为准),过期需重新调用下载接口|
| **依赖**| 完全依赖 SmartLib API 和网络连接,离线不可用|

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 1034)May include surrounding context.

md
|------|------|
| **检索不到想要的论文怎么办?** | 1. 去掉过滤条件扩大范围 2. 尝试上位词 3. 用英文关键词在接口4再试 4. 用 `U=` 替代 `T=` |
| **全文下载失败怎么办?** | 仅中文期刊支持全文下载。下载 URL 约 10 分钟有效(以下游返回为准),过期需重新调用。英文文献自动走多渠道下载策略获取 OA 版本。 |
| **Token 多久过期?** | Access Token 约 30 秒,Refresh Token 约 2 小时(以 API 返回为准)。系统自动管理刷新,用户无感知。 |
| **英文文献能不能下全文?** | 本技能集成十级多渠道下载策略(ArXiv → Unpaywall → CORE → OpenAlex 等),Gold/Green/Hybrid OA 论文成功率 >85%。付费墙内论文无法获取。 |
| **配额耗尽后还能用吗?** | 不能。试用额度耗尽后 Gateway 返回 429 拒绝所有检索请求,并自动弹出套餐选择;如需更高额度,回复套餐数字或说「升级 / 充值」即可(企业 / 机构定制联系 vipsmart@vipslib.com)。 |
| **计费 token(consume_token)是一次性的吗?** | **是的,务必注意。** 每次 `/consume` 返回的 `consume_token` **仅能使用一次**且 **约 60 秒过期**。获取后要立刻调用 `/search` 或 `/download`,不要缓存或复用;过期/已用需重新 `/consume`。这与 SmartLib 云端 Access/Refresh Token(约 30s/2h,以 API 返回为准,系统自动刷新)是两回事。 |

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 1066)May include surrounding context.

md
- 检索策略遵循三级分级体系:默认平衡策略,综述自动切换宽检索,引用自动切换窄检索
- 检索→评估→调整→再检索是核心工作流
- Access Token 有效期约 30 秒,Refresh Token 约 2 小时(以 API 返回为准),系统自动管理刷新
- 全球文献检索(接口4)仅提供元数据,部分无全文
- 中文期刊(接口1-3)支持全文下载,是核心优势,应优先推荐
- PageSize 建议不超过 100

The prescribed workflow sends research queries and article identifiers to external services without a privacy warning.

Medium
Category
Output Handling
Confidence
85% confidence
Finding

The agent workflow instructs external searches and DOI-based retrieval, with earlier sections routing searches through SmartLib Gateway and DOI lookups through Unpaywall. Although the architecture names these services, the guide does not warn users that their queries and selected article identifiers are disclosed externally and may reveal sensitive research interests.

Content

No source excerpt is available for this finding.

Shadow Command Trigger

Medium
Category
Trigger Abuse
Confidence
70% confidence
Finding

Shadow Command Trigger: description clause '"写文献综述""找几篇引用支撑论点"。英文:"find papers" "search literature" "write literature review" "find supporting citations"。

调用前必须先用 /consume 获取 consume_token,再凭 token 调 /search(每次计费接口调用都需一次 /consume)。' conflicts with built-in command 'search'

Content

No source excerpt is available for this finding.

Automatic web-based keyword expansion lacks a user-facing privacy disclosure.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The skill automatically uses web searches to expand user-supplied research terminology, sending those terms to additional providers beyond the selected literature service without a user-facing disclosure or sensitivity check. For confidential research, patent development, or unpublished projects, this can expose proprietary topics through an ancillary operation the user did not explicitly request.

Content

No source excerpt is available for this finding.

The template explicitly requires Chinese plan names and Chinese status mappings without user language choice.

Low
Category
Output Handling
Confidence
92% confidence
Finding

Line 17 requires a Chinese plan name, and line 29 specifies Chinese status mappings in the AI conversation. These are explicit language requirements without user opt-in or a documented justification for restricting the output language.

Content

No source excerpt is available for this finding.