The Agent Skills Directory

[COMMAND_EXECUTION]: The RedisCacheAdapter implementation provided in the examples uses the PHP unserialize() function to process data retrieved from the cache.
Evidence: Found in references/examples.md at the line return unserialize($value);.
Risk: This is a classic unsafe deserialization vulnerability. If an attacker can manipulate the data stored in the Redis cache, they can trigger PHP Object Injection, which often leads to arbitrary code execution.
[PROMPT_INJECTION]: The skill functions by interpolating user-controlled variables directly into PHP code templates, creating an indirect injection surface.
Ingestion points: Template placeholders such as {Name}, {operation}, {params}, and {returnType} defined in SKILL.md and references/templates.md.
Boundary markers: The skill does not define specific delimiters or instructions to the agent to ignore malicious patterns within these inputs.
Capability inventory: The skill is designed to generate (and potentially write) multiple PHP files to the local file system (e.g., src/Domain/, src/Infrastructure/).
Sanitization: No validation or sanitization rules are specified for the placeholder values, allowing for the potential injection of arbitrary PHP logic into the resulting codebase.

create-adapter