wb-xhs-humanize-compliance
Fail
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill includes instructions in the 'Runtime bootstrap and update gate' section to 'automatically install or update a missing or outdated dependency' and 'run its diagnostic'. This pattern allows the agent to execute arbitrary code during the skill's bootstrap phase based on an external policy file.
- [EXTERNAL_DOWNLOADS]: The instructions reference an external GitHub repository ('github.com/yuwen-cool/yuwen-publish-precheck') as a source for its 'publish-precheck' logic and methodology.
- [COMMAND_EXECUTION]: The bootstrap process mandates the execution of shell commands to manage software packages and verify installation through diagnostic checks.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to ingest and process untrusted user-supplied content (AI-generated drafts) while possessing system-level execution capabilities.
- Ingestion points: User-provided Xiaohongshu drafts processed in SKILL.md.
- Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore embedded instructions within user drafts.
- Capability inventory: The agent is authorized to perform package installations and diagnostic executions through the runtime update policy.
- Sanitization: No sanitization or filtering logic is specified for the ingested content.
Recommendations
- AI detected serious security threats
Audit Metadata