csm-sentiment-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to retrieve and analyze untrusted customer text, including email text, chat transcripts, and journal comments. This creates an indirect prompt injection surface where adversarial content embedded within customer messages could attempt to influence the agent's analysis or output formatting.
- Ingestion Points: Untrusted content enters the context at Step 2 (
sys_journal_field), Step 3 (sys_email), and Step 4 (interaction_entry). - Boundary Markers: Explicit delimiters or instructions telling the model to ignore executable prompts inside customer text are absent.
- Capability Inventory: The skill is low-risk as its tools are limited to reading data and rendering a structured markdown text report, lacking dangerous capabilities like system file modification or outbound internet execution.
- Sanitization: No pre-processing or sanitization rules are specified for user-supplied message inputs before analysis.
Audit Metadata