The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill requires cloning a repository from an untrusted source (https://github.com/webzler/agentMemory.git). This source is not on the trusted list and lacks verification.
[COMMAND_EXECUTION] (HIGH): The installation process involves running npm install and npm run compile on the untrusted code, followed by npm run start-server. This provides a path for arbitrary code execution on the host machine.
[PROMPT_INJECTION] (HIGH): The skill implements a persistent memory system that is highly vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Untrusted data enters the agent context through the memory_write tool, which stores strings from any source the agent processes (e.g., external web content).
Boundary markers: Absent. There are no instructions or delimiters to prevent the agent from obeying commands stored within the memory content.
Capability inventory: The skill has file-write capabilities (syncing with documentation) and persistent storage. Maliciously crafted memories can influence any future agent decision that relies on memory_search or memory_read outputs.
Sanitization: Absent. There is no evidence of validation or filtering for executable instructions within stored memories.

agent-memory-mcp