The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill is designed to cache 'expensive API results' and 'search results' (Category 8c: Tool output poisoning).
Ingestion points: Data enters the system via the set command in scripts/cache_manager.py (lines 191-201).
Boundary markers: None identified. The script stores raw string or JSON data without sanitization or injection-resistant wrapping.
Capability inventory: While this specific script only performs Redis operations, the SKILL.md (lines 4-5) explicitly states it is used to 'save context' for 'agent sessions'. This creates a high-risk surface where an agent might read and trust data from the cache that originated from an untrusted external API.
Sanitization: No sanitization of the value is performed before storage or after retrieval.
[Credential Exposure] (MEDIUM): The script scripts/cache_manager.py (lines 35-43) reads REDIS_PASSWORD from environment variables. While this is standard practice, the use of a .env file increases the risk of accidental exposure if the environment is not properly secured.
[Command Execution] (LOW): The skill documentation describes a wrapper script scripts/cache (not provided in the files) that 'creates a venv and installs dependencies' (SKILL.md line 12). Automated environment setup can sometimes be leveraged for unintended execution, though it is a common convenience feature.

memory-cache