A merged PR adds four Ejentum cognitive-harness skills to the Claude Code plugin registry, enabling the agent to call an external MCP server for structured reasoning scaffolds before generating code or architectural plans.
Integrates an external MCP-based reasoning harness that injects explicit failure patterns, procedural steps, and suppression vectors into Claude Code's context before generation, forcing a deliberate planning phase for code refactoring and architecture tasks.
Developers using Claude Code can get more deliberate and less error-prone code generation because the agent now runs structured reasoning checks before writing output. By routing prompts through the Ejentum MCP server, the plugin injects explicit cognitive scaffolds into the agent's context, which reduces impulsive refactoring mistakes and architectural hallucinations during complex multi-file edits. Teams should monitor whether the external MCP call adds noticeable latency to interactive coding sessions and verify that the injected scaffolds consistently improve output quality rather than simply consuming context window tokens without measurable gains.