a1-yandex-kit
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes store data such as delivery notes, order comments, and customer names, which are written by third parties. This creates an attack surface for indirect prompt injection where malicious instructions could be embedded in the API data to influence agent behavior.\n
- Ingestion points: Free-text fields in store data (delivery notes, order comments, customer names/notes, product descriptions/reviews) are ingested and used as task-relevant input (SKILL.md).\n
- Boundary markers: The skill provides clear instructions for the agent to ignore instructions embedded in store text and identify the specific object and field if content must be reported (SKILL.md).\n
- Capability inventory: The agent has access to tools for modifying store state, including catalog management, order processing, and promotion configuration via the Yandex KIT API tools and Bash scripts (SKILL.md).\n
- Sanitization: No programmatic sanitization or filtering is implemented; the skill relies on reasoning boundaries and instructional constraints provided in the SKILL.md file.
Audit Metadata