The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions include running npm install and npm run start-server, which executes code from a third-party community repository (webzler/agentMemory).
[EXTERNAL_DOWNLOADS]: The skill downloads and installs dependencies from a community-managed GitHub repository (https://github.com/webzler/agentMemory/) which does not have a verified provenance.
[PROMPT_INJECTION]: The skill defines a memory layer that imports context from .kilocode/, .clinerules/, and .roo/ directories, creating an attack surface for indirect prompt injection.
Ingestion points: Processes markdown files within project-level configuration folders.
Boundary markers: No specific delimiters or warnings to ignore embedded instructions are provided.
Capability inventory: Uses MCP tools to read and search this data, influencing agent logic.
Sanitization: No sanitization of ingested markdown content is documented.
[COMMAND_EXECUTION]: The server startup command npm run start-server <project_id> <absolute_path_to_workspace> incorporates user-controlled variables which could be leveraged for command injection if inputs are not validated.
[PROMPT_INJECTION]: The skill includes a risk: critical field in its metadata, which is a deceptive indicator that attempts to pre-emptively define the skill's security status.

agent-memory