The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The setup instructions require cloning a repository from an external, unknown source (https://github.com/webzler/agentMemory.git) into the agent's local environment.
[REMOTE_CODE_EXECUTION]: Following the download, the skill instructs the execution of npm install and npm run compile on the external codebase, which triggers the execution of scripts and dependency installations from that untrusted source.
[COMMAND_EXECUTION]: The skill runs a local MCP server and a dashboard using npm run, which executes scripts defined in the external repository with access to the target workspace path provided by the user.
[INDIRECT_PROMPT_INJECTION]: The skill provides tools to search and read from a memory bank synced with project documentation. If a malicious actor places instructions within the project files being indexed, the agent may retrieve and execute these instructions during memory retrieval tasks.
Ingestion points: The memory_read and memory_search tools read data from the local memory bank (defined in SKILL.md).
Boundary markers: None present; there are no instructions to the agent to disregard formatting or potential commands within the retrieved memory content.
Capability inventory: The skill possesses filesystem read/write capabilities and the ability to start local servers via the MCP interface.
Sanitization: No evidence of sanitization, filtering, or validation of the ingested memory content is provided.

agent-memory-mcp