redis
Fail
Audited by Gen Agent Trust Hub on Feb 14, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection] (HIGH): The skill processes untrusted data from an external source (Redis) without boundary markers or sanitization.
- Ingestion points: Data retrieved from Redis via
redis-cli GETandredis-cli KEYS(SKILL.md). - Boundary markers: None present to differentiate database content from agent instructions.
- Capability inventory: Execution of
redis-clicommands for reading, writing, and flushing data (SKILL.md). - Sanitization: No evidence of input filtering or output escaping.
- [Command Execution] (HIGH): The skill utilizes the
redis-clibinary and requirescurlandjqto be available on the host system (SKILL.md), posing a risk of arbitrary command execution if the agent is manipulated. - [Data Exposure] (MEDIUM): Interaction with database keys can expose sensitive credentials, session tokens, or private data stored in the memory cache (SKILL.md).
Recommendations
- AI detected serious security threats
Audit Metadata