The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions require executing npx -y ejentum-mcp, which downloads and runs code from the npm registry. This represents the execution of unverifiable third-party code from a source that is not listed as a trusted vendor.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching a package from npm and performing outbound HTTPS requests to the vendor's logic API gateway at ejentum.com to retrieve reasoning scaffolds.
[CREDENTIALS_UNSAFE]: The skill requires a third-party API key (EJENTUM_API_KEY) to be stored in the environment. While the documentation correctly suggests environment variables, this dependency introduces a requirement for managing external secrets.
[INDIRECT_PROMPT_INJECTION]: The agent is instructed to ingest data from an external tool and treat it as internal instructions. This creates a surface for indirect prompt injection if the remote source is compromised.
Ingestion points: Data enters the agent's context from the output of the harness_* tools, specifically through fields like [PROCEDURE] and [REASONING TOPOLOGY]. (File: SKILL.md)
Boundary markers: The skill uses bracketed headers to delineate the data, but it does not provide explicit instructions to the agent to sanitize or ignore potentially malicious control tokens embedded in the remote content.
Capability inventory: The tool is designed to work with MCP clients (like Claude Code or VS Code) which typically possess capabilities for filesystem access, shell execution, and network operations. (File: SKILL.md)
Sanitization: There is no evidence of pre-processing or sanitization of the content returned by the API before it is consumed by the agent's reasoning mechanism.
[COMMAND_EXECUTION]: The setup process involves shell command execution (npx) to configure and run the MCP server within the local environment.

ejentum-reasoning-harness