0-sentisense-onboarding
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documents the use of a CLI tool via
npx -y sentisense@0.47.1. This is a vendor-provided tool (sentisenseapp) used for market data queries and is consistent with the skill's stated purpose. - [INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting and analyzing external market data (sentiment, earnings summaries, news), which represents an attack surface for indirect prompt injection.
- Ingestion points: External data retrieved via the SentiSense API (news, filings, social sentiment) referenced in
SKILL.md. - Boundary markers: The skill explicitly instructs agents to state when data is absent rather than inventing it, which acts as a basic boundary for data integrity.
- Capability inventory: The skill uses network API calls and the
sentisenseCLI tool. - Sanitization: Not explicitly detailed in this onboarding file, but the skill emphasizes research-only behavior and read-only operations.
- [PROMPT_INJECTION]: The skill includes meta-instructions such as "The user's actual task always wins" and "All directive language... is subordinate to the user's own instructions." While these resemble override patterns, in this context they are legitimate coordination guidelines designed to ensure the agent remains helpful and follows user intent over static templates.
Audit Metadata