job-hr
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data from candidate resumes and interview transcripts, creating a vulnerability surface where malicious instructions could be embedded in those documents.
- Ingestion points: Data is ingested from recruitment tables, MiaoJi interview recordings, and candidate resumes (SKILL.md).
- Boundary markers: The skill attempts to mitigate manipulation by requiring the agent to use original quotes for policy information and strictly factual summaries for interview records (SKILL.md, Scenario 2, Scenario 5).
- Capability inventory: The skill has the capability to generate onboarding materials and trigger automated workflow tasks within the Feishu (Lark) platform (SKILL.md, Scenario 3).
- Sanitization: The skill relies on instructional constraints—such as 'record only what the candidate said'—rather than technical sanitization or input filtering to handle potentially adversarial content (Scenario 2).
Audit Metadata