Salary observation to tax calculator
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to instruct the agent on how to ingest data from a previous tool's observation and use it in a subsequent tool call, which creates a vulnerability surface.
- Ingestion points:
rules.mdandsolver_rules.mdspecify extracting salary data from an<obs>observation block in the dialogue history. - Boundary markers: The instructions rely on
<obs>tags as markers for tool outputs. - Capability inventory: The skill utilizes the
TaxCalculatortool call, which accepts the extracted data as a parameter, potentially allowing malicious tool output to influence the next system action. - Sanitization: No explicit sanitization or validation of the ingested content is defined in the instructions, although the
validator.pyscript ensures the parameter is grounded in the observation.
Audit Metadata