goal-os-optimize

Automate agent OS maintenance by invoking the /goal command to trigger a self-directed optimization loop that audits skills, CLAUDE.md files, rules, and project catalogs. It coordinates a primary agent and a separate judge LLM to validate iterations and ensure governance compliance.

The optimization loop manages skills and project catalogs by running five modes: clean, sharpen, revive, forge, and maintain. It audits and prunes duplicate skills, aligns CLAUDE.md rules, and catalogs projects with guardrails to reduce manual cleanup time.

Start optimizing CLAUDE.md rules and skill inventories by invoking the /goal command with a concise objective. This initiates the optimization loop, requiring a primary agent to execute changes and a judge agent running on a separate LLM to validate them within a finite iteration budget.

Yes, you can use a separate LLM to validate iterations during agentic OS optimization. The workflow requires a judge agent running on a distinct LLM to evaluate the primary agent's changes, ensuring compliance with stated goals and maintaining governance across the workspace.

Limitations of this automated loop include its dependency on a finite iteration budget and the requirement for a separate LLM to run the judge agent. It is designed for ongoing maintenance workflows across skill libraries and project inventories rather than one-off fixes.

The best way to prune duplicate skills and align rules is to use the clean and sharpen modes within the optimization loop. These modes audit skill libraries and CLAUDE.md files, applying guardrails to catalog projects and reduce manual governance overhead.