create-or-audit-skill

To create codebase-specific skills, you must reference at least 3 real file paths, commands, or patterns unique to the aihub-core repository, include concrete verification steps, and avoid generic programming advice to ensure portability and quality.

Auditing existing skills requires a 6-gate review protocol to identify generic, non-repo-specific skills that waste AI context window space, checking for portability, redundancy, and concrete verification steps before revising or removing them.

Generic skills fail to capture unique team workflows because they lack references to real codebase-specific file paths and commands, leading to inconsistent processes, broken automations, and wasted AI context window space across the repository.

Yes, you can build a custom skill to encode exact deployment steps for a new LiteLLM provider integration, including required config file edits, registration steps, and validation commands so every engineer follows the same proven process.

To ensure a Dagster pipeline workflow skill fails outside your codebase, you must reference at least 3 real file paths, commands, and patterns unique to the aihub-core repository, and include concrete validation checks rather than generic programming advice.