Security audit

Discharge Summary Writer

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent discharge-summary purpose, but it can overwrite arbitrary writable files and can let untrusted patient text alter the generated clinical document.

Install only if you will run it in a sandbox or dedicated workspace, review every generated summary with a qualified clinician, avoid passing untrusted patient text directly into the generator, and constrain output paths so the script cannot overwrite unrelated files. Treat the language setting as unreliable until English rendering or explicit locale handling is fixed.

A.I.G

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 (2)

T09 · Insecure Skill Coding Practices

Warning

Location: scripts/main.py:315
Finding: Unrestricted Output Path Allows Arbitrary Writable-File Overwrite
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Warning

Location: scripts/main.py:42
Finding: Unescaped Patient Data Permits Markdown and Raw HTML Content Injection
Content: View full analysis
str: """Format patient information section.""" return f"""## 患者基本信息 / Patient Information | 项目 / Item | 内容 / Value | |------------|-------------| | 姓名 / Name | {info.get('name', 'N/A')} | | 性别 / Gender | {info.get('gender', 'N/A')} | | 年龄 / Age | {info.get('age', 'N/A')} 岁 | | 住院号 / MRN | {info.get('medical_record_number', 'N/A')} | | 入院日期 / Admission Date | {info.get('admission_date', 'N/A')} | | 出院日期 / Discharge Date | {info.get('discharge_date', 'N/A')} | | 住院天数 / Length of Stay | {info.get('hospital_stay_days', 'N/A')} 天 | | 科室 / Department | {info.get('department', 'N/A')} | | 主管医师 / Attending Physician | {info.get('attending_physician', 'N/A')} | """ ``` The same direct interpolation pattern is also used for clinical history, diagnoses, procedures, findings, complications, consultations, medications, appointments, and warning signs throughout lines 59–193. ### Technical Analysis Patient-controlled JSON values are inserted directly into Markdown without context-sensitive escaping or validation. The validation routine checks only for the presence of a limited set of fields; it does not verify field types, lengths, control characters, Markdown syntax, or raw HTML. An attacker can therefore provide values containing: - New headings or sections. - Table delimiters and line breaks that escape the intended table cell. - Forged diagnoses, prescriptions, follow-up instructions, or signature content. - Raw HTML accepted by permissive Markdown renderers. - Links or image syntax referencing remote resources. This is particularly significant because the output is a clinical document. Injected formatting can make attacker-controlled content appear to be legitimate medical guidance. If ...[truncated 2047 chars]
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (8)

Undeclared Tool Scope

Medium

Category: MCP Least Privilege
Confidence: 86% confidence
Finding: The skill documents file read/write behavior via a local Python script but does not declare any explicit tool scope such as permissions or allowed-tools. In a clinical context handling PHI, undeclared filesystem capability increases the risk of unintended access to sensitive files, path misuse, and weaker policy enforcement by the hosting agent platform.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The trigger description is broad enough to activate on general clinical documentation requests, not just discharge summaries. Over-broad routing can cause the wrong skill to process requests containing sensitive patient data, leading to inappropriate document generation, data mixing, or use outside the intended clinical workflow.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: Defaulting output language to Chinese without explicit user selection can generate discharge instructions in a language the clinician or patient did not request. In medical discharge documentation, this creates patient-safety and compliance risk because medication instructions, follow-up plans, and warning signs may be misunderstood or unusable.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 81% confidence
Finding: This JSON reference file contains user-facing natural-language medical content exclusively in Chinese, including instructions and warning signs. Because the file does not document that the skill is region-specific or that users can opt into this language, it may violate the policy against forcing a specific language without user choice.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: This JSON maps medical terms to Chinese text throughout the file, effectively enforcing a specific language/locale in the skill data. The file does not indicate that Chinese localization is optional, user-selected, or limited to a justified region-specific use case.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: This markdown file contains natural-language guidance and sample content written in Chinese, but it does not state that language selection depends on user preference or context. Because SQP-3 applies to all file types and covers forced language/locale behavior, these examples may imply a default output language without opt-in.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The CLI explicitly offers --language choices of zh and en at L280-L285, but generate_structured_output hardcodes 'zh' and generate_summary itself does not vary output by the language parameter. This creates a natural-language policy issue because the skill appears to offer language choice while effectively enforcing Chinese output regardless of user preference.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 78% confidence
Finding: The document title is fixed as "出院小结 / DISCHARGE SUMMARY," and many section labels throughout the template are presented in Chinese alongside English without any indication that language selection is optional. This can be a natural-language locale policy issue because the template appears to prescribe a specific bilingual format rather than allowing user or organizational language preference.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.