lark-attendance
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data retrieved from the Lark API, which presents a surface for indirect prompt injection if attendance records or exception explanations contain malicious instructions.
- Ingestion points: Data enters the context via the
lark-cli attendance user_tasks querycommand output in the '查询工作流' (Query Workflow) section. - Boundary markers: The skill uses a structured '可追溯输出模板' (Traceable Output Template) to delimit agent output and prevent the interpolation of raw, unverified data.
- Capability inventory: The skill uses
lark-clifor read-only schema inspection and data retrieval. No write operations or system-level modifications are enabled. - Sanitization: The instructions explicitly mandate data desensitization ('输出已脱敏'), forbidding the inclusion of access tokens, IDs, or full employee data in the final report. It also instructs the agent to report raw findings rather than interpreting or 'guessing' reasons for missing records, which reduces the risk of obeying embedded instructions.
Audit Metadata