byted-recruitment-bosszhipin-self-submission
Pass
Audited by Gen Agent Trust Hub on Apr 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes shell commands for file system organization and clipboard operations.
- Evidence: Use of
mkdir -pto create position-specific directories. - Evidence: Use of
xclip -o > [filename]to capture resume text from the clipboard. - Evidence: Instructions for the agent to configure
cronscheduled tasks for periodic execution. - [DATA_EXFILTRATION]: The skill collects candidate PII (resumes) and sends data to an external communication platform.
- Evidence: Candidate information and processing summaries are synchronized to a Lark/Feishu channel.
- [PROMPT_INJECTION]: The skill processes untrusted external content from resumes and job descriptions, introducing a surface for indirect prompt injection.
- Ingestion points: External content from BOSS Zhipin resumes and JDs (SKILL.md).
- Boundary markers: Absent; content is copied directly into local markdown files.
- Capability inventory: Shell execution (
mkdir,xclip), local file writing, and network synchronization to Lark. - Sanitization: None; the skill captures all clipboard content directly without filtering.
Audit Metadata