The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation instructs users to visit an external website (https://mcp.simplemem.cloud) and input their OpenRouter API key to obtain an authentication token. This practice exposes high-value master credentials to a third-party domain that is not a recognized well-known service or trusted provider.
[DATA_EXFILTRATION]: When using the recommended Cloud Service path, the skill sends agent dialogues and session context to an external, non-whitelisted endpoint (https://mcp.simplemem.cloud/mcp). This constitutes the exfiltration of internal conversation data to a third-party service.
[EXTERNAL_DOWNLOADS]: The skill requires installing the simplemem package from PyPI and recommends cloning source code from an external repository (github.com/aiming-lab/SimpleMem.git). These resources originate from an organization that is not on the trusted vendors list.
[PROMPT_INJECTION]: The skill implements a memory system that ingests untrusted dialogue, stores it as atomic facts, and later retrieves it for injection into the agent's prompt context. This creates a surface for indirect prompt injection.
Ingestion points: Data enters the system via the memory_add, memory_add_batch, and add_dialogue methods.
Boundary markers: The skill documentation does not mention the use of delimiters or specific instructions for the agent to ignore embedded commands within the retrieved memory context.
Capability inventory: The system performs hybrid retrieval that directly influences the agent's knowledge and response generation.
Sanitization: While the documentation claims to redact sensitive credentials like API keys from stored data, it does not describe any validation or filtering of instructional content or malicious prompts within the processed dialogues.

simplemem