The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface.
Ingestion points: The skill is instructed to read ../../CLAUDE.md for scenario configuration and fetch real-time financial metrics from external [ERP] and [BI] systems.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between the skill's logic and the data retrieved from external sources.
Capability inventory: No high-risk capabilities such as arbitrary command execution, file writing, or network exfiltration were detected in the skill instructions.
Sanitization: The skill lacks any instructions for validating, escaping, or filtering data ingested from the external ERP/BI systems or the configuration file before it is used to generate the dashboard.

kpi-dashboard