Ae1
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md 默认再用 `scripts/render_report.py` 生成一份可分享的 HTML 版报告并告知保存路径(详见 report-template.md「HTML 报告导出」)。
Security audit
Security checks for vulnerabilities and agentic risk
The skill is a coherent company bidding-intelligence integration that discloses its API use, report generation, onboarding flow, and local credential storage, though users should understand the privacy and persistence tradeoffs before using auto-registration.
Install only if you are comfortable using Zhiliaobiaoxun's API, spending account credits for reports, and allowing optional first-use registration to send a hashed MAC-derived device identifier. Preconfigure ZLBX_API_KEY if you want to avoid auto-registration and device-feature collection.
Referenced artifact was not completely inspected
默认再用 `scripts/render_report.py` 生成一份可分享的 HTML 版报告并告知保存路径(详见 report-template.md「HTML 报告导出」)。
Referenced artifact was not completely inspected
默认再用 `scripts/render_report.py` 生成一份可分享的 HTML 版报告并告知保存路径(详见 report-template.md「HTML 报告导出」)。
This duplicate finding reflects the same hidden-instruction issue: behavior-affecting guidance is concealed inside an HTML comment rather than exposed in normal skill metadata or visible documentation. Even if intended for package bookkeeping, concealed directives make adversarial instruction stuffing easier and hinder reliable auditability.
安装包身份与升级检查:所有发往知了标讯 API 的请求(包括账户查询)必须额外携带以下两个请求头:
X-Skill-Package: bid-award-competitive-analysis/1.0.5
This duplicate finding reflects the same hidden-instruction issue: behavior-affecting guidance is concealed inside an HTML comment rather than exposed in normal skill metadata or visible documentation. Even if intended for package bookkeeping, concealed directives make adversarial instruction stuffing easier and hinder reliable auditability.
安装包身份与升级检查:所有发往知了标讯 API 的请求(包括账户查询)必须额外携带以下两个请求头:
X-Skill-Package: bid-award-competitive-analysis/1.0.5
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
The file begins with embedded hidden operational instructions that tell the agent which identity headers to attach and how to behave across requests. Hidden control-plane instructions inside skill content are dangerous because they can override user expectations, conceal external attribution behavior, and make it harder for reviewers to distinguish harmless content from privileged execution directives.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
The file begins with embedded hidden operational instructions that tell the agent which identity headers to attach and how to behave across requests. Hidden control-plane instructions inside skill content are dangerous because they can override user expectations, conceal external attribution behavior, and make it harder for reviewers to distinguish harmless content from privileged execution directives.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
This duplicate finding flags the same hidden instruction block that attempts to control request identity and API usage from within the template content. Even if the values are not themselves secret, allowing hidden prompt content to dictate credential-coupled behavior creates a prompt-injection style control channel over authenticated requests.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
This duplicate finding flags the same hidden instruction block that attempts to control request identity and API usage from within the template content. Even if the values are not themselves secret, allowing hidden prompt content to dictate credential-coupled behavior creates a prompt-injection style control channel over authenticated requests.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
This duplicated finding identifies the same hidden-instructions issue: concealed markdown comments are being used to direct authenticated API behavior and package identity handling. Because the content is treated as model-readable instructions rather than runtime-enforced configuration, an attacker or compromised package can influence how the agent uses credentials and external services without sufficient transparency or isolation.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
This duplicated finding identifies the same hidden-instructions issue: concealed markdown comments are being used to direct authenticated API behavior and package identity handling. Because the content is treated as model-readable instructions rather than runtime-enforced configuration, an attacker or compromised package can influence how the agent uses credentials and external services without sufficient transparency or isolation.
请求身份:使用已配置的 API Key,并保留本包的 X-Client,携带下列完整身份头;不得沿用其它包的版本或来源。
X-Client: company-intel/1.0.2
Code contains obfuscation (base64, hex encoding with execution). This is often used to hide malicious functionality.
# 知了标讯白色 logo(299x96 PNG base64 内嵌,保证报告离线/打印/转发时不裂图)
_LOGO_B64 = "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
...[truncated 27 chars]
The skill explicitly states it will write HTML reports to a local directory, but it does not declare any tool scope or permissions boundary for file writing. That mismatch weakens least-privilege controls and can let the skill write files in environments where operators cannot clearly audit or constrain that behavior.
The activation language is very broad and says the skill must be used even when users only generally want to understand a company through bidding records. Overbroad routing can cause the agent to invoke a networked, credit-consuming skill in situations that do not require it, increasing risk of unnecessary external data sharing, unexpected charges, and scope creep.
The skill explicitly instructs the agent to return project contact phone numbers 'as-is', including potentially full personal phone numbers from a paid account, while providing no meaningful privacy, consent, purpose-limitation, retention, or anti-abuse safeguards. In a competitive-intelligence workflow, this creates a real risk of exposing personal contact information for unsolicited outreach, profiling, or bulk harvesting.
The file embeds a full account lifecycle workflow—auto-registration, API key acquisition, persistence, and recharge/login handling—that materially exceeds the declared bidding-analysis purpose of the skill. This scope expansion creates unnecessary authority and data-handling paths, increasing the chance of unauthorized account actions and covert credential management under the guise of an analysis tool.
The document provides fixed Chinese wording to present to the user and repeatedly instructs the agent to output specific Chinese text and links. Because this file is a markdown skill instruction and does not offer any user language choice or justify a Chinese-only locale constraint, it conflicts with the language/locale policy described for SQP-3.
The skill instructs the agent to collect device fingerprints, including a MAC-derived hash, for trial-account creation even though that data is unrelated to bidding analysis. Collecting and transmitting stable device identifiers expands privacy risk, enables cross-session/device tracking, and normalizes hidden telemetry in a context where users would not reasonably expect it.
This section explicitly instructs the agent to send collected device features to an external auto-registration endpoint. Because the transmitted data includes persistent device-derived identifiers and triggers account creation, it represents an external data exfiltration path not necessary for the advertised analysis task.
> ### ⚠️ 请求体必须用 JSON 序列化函数生成,不要手拼字符串
>
> 用 `json.dumps(payload)` / `requests.post(url, json=payload)` / `JSON.stringify(payload)`,
> 或 `curl -d @file`;**不要用字符串拼接,也不要用 Python 的 `str(dict)`**
> (后者产出单引号,服务端会报 `Expecting property name enclosed in double quotes`)。
>
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
> ### ⚠️ 请求体必须用 JSON 序列化函数生成,不要手拼字符串
>
> 用 `json.dumps(payload)` / `requests.post(url, json=payload)` / `JSON.stringify(payload)`,
> 或 `curl -d @file`;**不要用字符串拼接,也不要用 Python 的 `str(dict)`**
> (后者产出单引号,服务端会报 `Expecting property name enclosed in double quotes`)。
>
> 历史教训:曾有版本采集 `home_path`,Windows 的 `C:\Users\alice` 直接拼进 JSON 字符串时
The skill directs the agent to persist obtained API keys into the user's home-directory configuration and to merge with existing config, giving the skill account-management and credential-storage behavior outside its stated purpose. Storing secrets from within a content skill increases the blast radius of prompt injection or misuse and can silently alter long-lived authentication state on the host.
The skill instructs creation and modification of ~/.zlbx/config.json, including merging existing settings and immediately reusing the stored key in-session. This creates unauthorized session persistence and long-lived host state changes from within a skill, which is risky because an analysis-oriented skill should not silently establish durable authentication artifacts on the user's machine.
注意事项:
- 目录不存在时先 `mkdir -p ~/.zlbx`
- 文件已存在时**合并而非覆盖**(保留用户可能的其他配置)
- `source: "auto"` 字段必须写入,**这是后续判断「是否输出自动登录链接」的关键依据**
The skill instructs creation and modification of ~/.zlbx/config.json, including merging existing settings and immediately reusing the stored key in-session. This creates unauthorized session persistence and long-lived host state changes from within a skill, which is risky because an analysis-oriented skill should not silently establish durable authentication artifacts on the user's machine.
注意事项:
- 目录不存在时先 `mkdir -p ~/.zlbx`
- 文件已存在时**合并而非覆盖**(保留用户可能的其他配置)
- `source: "auto"` 字段必须写入,**这是后续判断「是否输出自动登录链接」的关键依据**
The file title and surrounding instructions define the report template entirely in Chinese and do not provide any user opt-in or alternative locale. This can violate language/locale policy because it forces a specific output language regardless of user preference.