cloud-native-knowledge
Warn
Audited by Gen Agent Trust Hub on Mar 17, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The PHP implementation for Redis caching (
RedisCacheAdapterinSKILL.md) utilizes theunserialize()function on data retrieved from an external backing service. - Evidence:
return $value !== false ? unserialize($value) : null;inSKILL.md. - Risk: This pattern is susceptible to PHP Object Injection if the Redis store is compromised or if an attacker can influence the cached data, potentially leading to arbitrary code execution.
- Mitigation: Recommended practice is to use
json_encode()andjson_decode()for data serialization to avoid instantiating arbitrary objects. - [PROMPT_INJECTION]: The skill provides patterns for code generation and auditing without explicit instructions for the agent to sanitize or ignore instructions embedded in the data it processes, creating an indirect prompt injection surface.
- Ingestion points: Processes architectural patterns and code snippets for auditing and generation.
- Boundary markers: Absent in the provided implementation examples.
- Capability inventory: Includes Docker execution commands and system auditing patterns (
Grep,Glob) for infrastructure management. - Sanitization: No mention of sanitizing external inputs or cache data before processing or interpolation into commands.
Audit Metadata