data-labeling
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external datasets (e.g., customer reviews or general text) for labeling and classification. This creates a surface where adversarial instructions embedded within the data could attempt to influence the agent's behavior.
- Ingestion points: Data is ingested from local files such as
reviews.csvor passed directly via Python DataFrames in the provided examples. - Boundary markers: No explicit delimiters or system instructions are provided to help the agent distinguish between labeling instructions and the data content itself.
- Capability inventory: The skill utilizes the
label-studio-sdkto interact with labeling environments and performs data analysis using standard libraries. It does not exhibit high-risk capabilities like arbitrary command execution or external network exfiltration. - Sanitization: There is no evidence of input sanitization or filtering to remove potential prompt injection vectors from the data being processed.
Audit Metadata