The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation in references/advanced.md provides an example of Redis caching that utilizes the unserialize() function on data retrieved from the cache.
Evidence: In references/advanced.md, the MyRedisService::getCachedData method calls unserialize($raw) on data fetched via a Redis connection.
Risk: Using unserialize() on data that could potentially be influenced by an attacker (e.g., if the Redis instance is shared or compromised) can lead to PHP Object Injection, which often results in remote code execution (RCE).
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the generation of Shopware 6 plugins, entities, and subscribers based on user-provided requirements, creating an attack surface where malicious input could influence the agent's output.
Ingestion points: User-provided descriptions of business logic, entity definitions, or service requirements in files like SKILL.md.
Boundary markers: The skill does not explicitly define or enforce boundary markers to separate user-provided content from the agent's internal instructions during the code generation process.
Capability inventory: The code generated using this skill has significant capabilities, including database access (DAL), file system operations (Flysystem), and network interactions (App System webhooks and Store API).
Sanitization: While the documentation in references/security.md correctly advises developers to sanitize output and avoid SQL injection in the resulting code, these measures do not protect the agent itself from being manipulated by the input descriptions.

shopware6