The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the user to install the ejentum-mcp package directly from the npm registry using npx.
[REMOTE_CODE_EXECUTION]: Execution of the command npx -y ejentum-mcp results in the download and execution of arbitrary code from a remote repository at runtime.
[COMMAND_EXECUTION]: The skill configuration requires adding a command to the local MCP settings (npx -y ejentum-mcp), allowing the agent to spawn processes on the host machine.
[DATA_EXFILTRATION]: The skill documentation confirms that the MCP server makes outbound HTTPS requests to an external API gateway (ejentum.com). While this is the intended functionality, it establishes a data flow from the agent's context to a third-party service not included in the trusted vendors list.
[CREDENTIALS_UNSAFE]: The skill requires an EJENTUM_API_KEY. While it correctly recommends using environment variables for storage, the key is transmitted to a third-party service for authentication.
[PROMPT_INJECTION]: The skill introduces a surface for indirect prompt injection (Category 8). It ingests user prompts to generate cognitive scaffolds.
Ingestion points: User prompts are passed to the ejentum harness tools.
Boundary markers: None identified in the skill instructions to separate user input from tool instructions.
Capability inventory: The skill possesses network capabilities to an external API.
Sanitization: No sanitization or validation of the external scaffold content is mentioned before it is 'absorbed' by the agent.

ejentum-reasoning-harness