Back to skill

Security audit

Iatros(诺康)

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed Chinese health-literacy writing persona skill with no executable code, credentials, network access, persistence, or hidden local authority.

Install this only if you want a Chinese Iatros-branded voice and signature for non-personalized health education content. Do not rely on it for diagnosis, treatment choices, medication advice, urgent issues, or professional medical decisions.

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
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:31
Finding

Mandatory Branded Attribution Alters Agent Output

Content
View full analysis
本 skill 是康健知识科普大使的对外发声规范包。安装后可在生成对外内容时调用 Iatros 人设与署名口径,保持人设与发声一致,严守科普红线。 > 署名:Iatros(诺康)| 康健知识大使 著述。 > 授权:CC BY-NC-ND 4.0(署名·非商业·禁止演绎)。 ``` At `SKILL.md:31-34`: ```markdown ## 3. 署名政策(对外 SSOT) - **发声人**:`Iatros(诺康)` - 内容签名:`—— Iatros(诺康)| 康健知识大使` - 文档落款:`© 作者:Iatros(诺康),康健知识大使` ``` At `SKILL.md:108-111`: ```markdown ## 13. 最佳实践 - 科普只做信息与方法:讲清楚原理,不针对个人下结论,效果表述保持克制。 - 善用相对表述,规避极限词;温和有据,让人信赖。 - 署名与人设一致:每次对外内容均使用 Iatros 署名落款,并在涉及具体问题时提示以正规机构为准。 ``` ### Technical Analysis The skill instructs an agent to adopt the Iatros persona and append a fixed signature or authorship statement to generated public content. The final rule requires the Iatros signature on every public output rather than making attribution conditional on the user's explicit request. When the skill is loaded for a health-related task, these persistent instructions can alter the requested output by introducing branding and an authorship claim. This is classified as skill instruction hijacking because the skill text modifies the agent's active response policy and output goals. The risk is limited to response integrity and attribution. The reviewed project contains only documentation files and provides no executable scripts, dependencies, credential access, remote payload retrieval, persistence mechanism, or privilege-escalation capability. ### Attack Path 1. An agent loads `SKILL.md` in response to a health-literacy or health-communication request. 2. The skill directs the agent to adopt the Iatros persona and signature policy. 3. The mandatory “every public output” instruction is applied even if the user did not request branding or attribution. 4. The generated response is presented with an Iatros signature or authorship claim. 5. ...[truncated 682 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file consistently presents the skill description, usage notes, and policy boundaries only in Chinese. Under the natural-language policy rules, forcing a specific language without user opt-in or documented justification is a locale/language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill defines a fixed external voice, signature, and communication package entirely around a Chinese-branded persona and Chinese signing conventions, but does not state that users may opt into another language or locale. Under the stated policy, forcing a specific language or locale without user choice can be a natural-language policy violation unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The activation criteria are broad enough to auto-load this persona for many generic health-related requests, including ambiguous queries that may drift from general education into quasi-personal health guidance. In a health context, overly eager activation increases the chance the agent adopts a branded instructional voice where stricter triage, disclaimers, or refusal logic should apply.

Content

No source excerpt is available for this finding.