The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external systems such as CRM, accounting, and payment records.
Ingestion points: Data is ingested from external connectors (~~бухгалтерия, ~~платежи, ~~crm) in SKILL.md.
Boundary markers: The instructions lack specific delimiters or warnings to ignore instructions that might be embedded within the business data (e.g., malicious payloads in transaction notes or customer records).
Capability inventory: The skill is allowed to use Bash, WebFetch, and Read tools, providing a significant surface for exploitation if an injection occurs.
Sanitization: There is no evidence of sanitization or filtering of the external data before it is interpolated into the narrative generation logic.
[DATA_EXFILTRATION]: The skill accesses and aggregates highly sensitive business information (revenue, margin, customer lists, and financial health) and exports it to external cloud storage providers like Yandex Disk and Mail.ru Cloud via the --save-to parameter.

quarterly-review