Back to skill

Security audit

sz-html

Security checks for vulnerabilities and agentic risk

Overview

This Chinese-language presentation skill uses local assets and user confirmations, with a packaging flaw relevant to redistribution.

Expect Chinese output and local HTML files. For other languages, request translated labels and matching HTML language markup. If rebuilding for redistribution, keep secrets out of the skill directory and never distribute a ZIP from a failed build.

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
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Credential Access

High
Category
Privilege Escalation
Confidence
96% confidence
Finding

The matched extension check is intended to prevent credential disclosure, not steal credentials, but it runs only after files have been written and the ZIP closed. Nonhidden .pem or .key files can therefore enter the distribution archive, and fail() leaves that sensitive archive on disk, where subsequent collection or distribution could expose it despite the failed build.

Content

Scanner excerpt · build.py (reported line 104)May include surrounding context.

python
zf.write(fp, arcname)
                added.append(arcname)
    leaked = [a for a in added
              if a.endswith(('.pem', '.key', '.env', '.zip')) or 'deploy_config' in a]
    if leaked:
        fail(f'交付包内出现敏感文件,已中止: {leaked}')
    size = os.path.getsize(zip_path)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Style-selection output is required to use Chinese names without user language opt-in.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

L211 explicitly requires selection cards to start with Chinese style names, and L81 also specifies Chinese color descriptions. These requirements impose Chinese on user-facing output without offering a language choice or documenting a region-specific justification.

Content

No source excerpt is available for this finding.

Style-selection cards are required to lead with Chinese names without a user language choice.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The instruction “中文名打头” explicitly requires Chinese names at the start of style-selection cards. No user opt-in or justification for this language constraint is provided in this source.

Content

No source excerpt is available for this finding.

Generated HTML is unconditionally assigned the zh-CN locale.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The accessibility requirements mandate lang="zh-CN" without checking the presentation's language or offering a locale choice. This imposes a specific locale and can cause assistive technologies to interpret non-Chinese content incorrectly.

Content

No source excerpt is available for this finding.

The preview generator fixes its output language and locale to Chinese without an opt-in.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The generated HTML unconditionally declares lang="zh-CN", alongside fixed Chinese interface text. The CLI offers no language or locale option, so users cannot choose an alternative output language.

Content

No source excerpt is available for this finding.