The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a long-term memory system where the agent can save and retrieve data across sessions. This persistence mechanism allows for the storage of potentially malicious instructions that could influence the agent when retrieved in a future context.\n
Ingestion points: Untrusted data enters the agent's context through the save_user_memory tool found in examples/memory_tools.py, which accepts a JSON string from the agent representing user facts or preferences.\n
Boundary markers: The get_user_memory tool retrieves and returns stored data without using explicit delimiters or instructions to the model to ignore any embedded commands within the memory content.\n
Capability inventory: The agent is granted capabilities to search, save, and delete data from its own long-term memory store.\n
Sanitization: The implementation in save_user_memory correctly validates that input is valid JSON but does not filter or sanitize the actual content for prompt injection patterns before storage.

agent-memory