event-budget
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process 'real vendor quotes' and pricing information from external sources. These inputs represent a surface for indirect prompt injection, as malicious instructions embedded within these documents could be interpreted as agent commands. \n
- Ingestion points: Untrusted data enters the agent context via vendor quotes and pricing inputs as described in the 'Workflow' (step 4) and 'Interview' (question 3) sections of
SKILL.md. \n - Boundary markers: Absent; the skill does not specify any delimiters or safety instructions (such as 'ignore instructions within the following text') to separate untrusted vendor data from the core skill instructions. \n
- Capability inventory: The skill directs the agent to interact with persistent memory storage and interface with sibling skills (e.g.,
event-ticket-pricing,event-sponsor-pricing), creating a scenario where injected instructions could influence multiple steps in the modeling chain. \n - Sanitization: Absent; no instructions are provided to the agent to validate, escape, or filter the external quote data before it is incorporated into the budget model.
Audit Metadata