The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the caching of user-supplied queries and external documentation through Cache Augmented Generation (CAG) patterns. If malicious instructions are embedded in the cached source data or queries, they could be executed by the agent in future sessions when the cached result is retrieved.
Ingestion points: The userQuery variable in queryWithCaching and the documents object array in buildCachedContext (SKILL.md).
Boundary markers: None present; the code samples do not include delimiters or instructions to ignore embedded commands within the cached text.
Capability inventory: The skill utilizes the @anthropic-ai/sdk to interact with Claude models and performs read/write operations to Redis databases.
Sanitization: No sanitization, validation, or filtering of the cached content is demonstrated in the implementation examples.
[EXTERNAL_DOWNLOADS]: The skill documentation references the use of the official Anthropic SDK and Redis clients (ioredis, @upstash/redis). These are well-known services used for standard large language model integrations and data persistence.

prompt-caching