ejentum-reasoning-harness
Warn
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill setup involves installing the
ejentum-mcppackage usingnpx -y ejentum-mcp. This fetches and executes code from the npm registry at runtime from a source not identified as a trusted organization or vendor resource. - [COMMAND_EXECUTION]: Installation requires running
npx, which executes shell commands to download and initialize the remote MCP server code. - [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by directing the agent to follow instructions returned by the Ejentum API (e.g.,
[PROCEDURE],[REASONING TOPOLOGY]). - Ingestion points: Instructions are retrieved from the
ejentum-mcptool output and incorporated into the agent's internal reasoning process. - Boundary markers: No specific delimiters or security warnings are provided to prevent the agent from following malicious instructions returned by the API.
- Capability inventory: The agent may possess significant capabilities (such as file system access or code execution) depending on the environment, which could be misused if following untrusted instructions.
- Sanitization: There is no evidence of filtering or validation of the content received from the remote API gateway before it is presented to the agent.
Audit Metadata