performance-hunter
Warn
Audited by Gen Agent Trust Hub on Jul 4, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The 'MultiLevelCache' pattern provided in
references/patterns.mdutilizesPickleSerializerfor storing data in Redis. Python'spicklemodule is inherently insecure because it can execute arbitrary code during deserialization. If an attacker manages to modify the data in the Redis cache, they could achieve remote code execution (RCE) on the host running the agent. - [PROMPT_INJECTION]: The skill's primary function is to analyze and profile user-provided code and performance data, which presents a surface for indirect prompt injection. \n
- Ingestion points: The skill ingests user code, SQL queries, and performance logs via the profiling and diagnosis instructions in
SKILL.md. \n - Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore embedded commands within the code or logs being analyzed. \n
- Capability inventory: The skill patterns involve executing database queries (
asyncpg), managing cache entries (redis), and performing network requests (aiohttp), providing significant capabilities for an injected instruction to exploit. \n - Sanitization: No sanitization or validation of the input data is suggested in the patterns or validation rules.
Audit Metadata