huashu-nuwa

Fail

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill explicitly instructs the agent to search for and download books from 'Z-Library' and 'LibGen'. These are well-known sources for pirated content that frequently host malicious files and are not considered trusted sources.
  • [COMMAND_EXECUTION]: The skill uses shell commands (bash and python3) to run local utility scripts stored within the skill's subdirectory (/scripts/). These scripts are invoked with parameters derived directly from external sources, such as <YouTube_URL>. If an attacker provides a crafted URL containing shell metacharacters (e.g., ;, &&, |), it could lead to arbitrary command execution on the host system.
  • [DYNAMIC_EXECUTION]: The core functionality of the skill is to dynamically generate and write new SKILL.md files to the .claude/skills/ directory. These generated files contain instructions that are subsequently executed by the AI agent, representing a form of dynamic code generation based on unvalidated external data.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits a high-risk surface for indirect prompt injection:
  • Ingestion points: The skill ingests untrusted data from multiple external sources, including web search results, YouTube transcripts, and user-provided PDFs in Phase 1.
  • Boundary markers: There are no explicit boundary markers or instructions to escape/sanitize the external data before it is interpolated into the newly generated SKILL.md file.
  • Capability inventory: The agent possesses significant capabilities, including file system write access to the .claude/skills/ directory, network access for research, and the ability to execute subprocesses.
  • Sanitization: No evidence of sanitization or validation for the content being processed into instructions was found.
  • [DATA_EXFILTRATION]: The skill performs directory scanning on .claude/skills/ to read existing skill descriptions. While intended for deduplication, this capability allows the skill to access and potentially expose the configuration or logic of other installed agent skills.
Recommendations
  • CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Sep 19, 2026, 11:32 PM
Security Audit — agent-trust-hub — huashu-nuwa